AI models guide · verified Sep 17, 2026 · 44 models · 6 providers · 11 model families · archive since 2026
Which AI should you use? Every model in PolyCog — what each is good at, how much it filters, and which few fit your project.
6 providers and 11 model families answer in PolyCog, and none of them is best at everything — and some will not touch some subjects at all. This page is the map: a strengths profile and a guardrails level for all 44 current models, a table of what each will and won’t do, a dated archive of every model that has left, and a box that reads what you’re working on and names the lineup to run. Nothing here calls out to a server — once loaded it works offline.
Try:
Runs in your browser. Nothing you type leaves this page. 343 project types in the dictionary.
The 6 seats at a glance
Each provider is one seat in a PolyCog debate. The Wildcard seat is one OpenRouter key that opens 6 more labs.
ChatGPT11 models
OpenAI’s line runs from a frontier flagship (GPT-6 Astra) through a mid tier built for everyday development to cheap minis for volume, with …
CodingReasoning & mathWritingModerate
Claude9 models
Anthropic’s line runs from a frontier flagship (Fable 5.1) through Opus and Sonnet to Haiku, which also runs PolyCog’s free preview. Anthrop…
CodingReasoning & mathWritingFirm
Gemini9 models
Google’s line pairs one Pro (3.1, still a preview id) with a deep Flash bench, and one Flash-Lite runs PolyCog’s free preview. Google positi…
CodingReasoning & mathLong documentsModerate
DeepSeek2 models
Two models: V4 Pro (visible chain-of-thought) and V4.1 Flash, which runs PolyCog’s free preview. DeepSeek positions the pair around reasonin…
CodingReasoning & mathSpeed & costLight
Grok3 models
Three models on one rate card, all with live web and X search. xAI positions the family around current information and a candid voice, and i…
CodingCurrent eventsAgentic workLight
Wildcard10 models
The sixth seat: one OpenRouter key opens Qwen, Kimi, GLM, Mistral, Llama and MiniMax — open-weight and non-US models, most of them very chea…
CodingReasoning & mathLong documentsguardrails vary by lab
Guardrails level, ranked
How much each model filters, from least to most. Five general dimensions make the level; a sixth — PRC-sensitive topics — is shown on its own because it moves independently of the rest. Scores are PolyCog’s dated read of each provider’s usage policy and observed behaviour in the app; they describe, they don’t referee, and nothing here is a guide to getting around a policy.
| GrokxAI · 3 models | Light · 2/5 | 1 | 2 | 2 | 2 | 1 | PRC topics: none |
| MistralMistral AI · Wildcard seat · 2 models | Light · 2/5 | 1 | 2 | 2 | 2 | 1 | PRC topics: none |
| DeepSeekDeepSeek · 2 models | Light · 2/5 | 2 | 2 | 2 | 2 | 1 | PRC topics: strict |
| GLMZ.ai · Wildcard seat · 2 models | Light · 2/5 | 2 | 2 | 2 | 2 | 1 | PRC topics: strict |
| KimiMoonshot · Wildcard seat · 2 models | Light · 2/5 | 2 | 2 | 2 | 2 | 1 | PRC topics: strict |
| MiniMaxMiniMax · Wildcard seat · 1 model | Light · 2/5 | 2 | 2 | 2 | 2 | 1 | PRC topics: strict |
| QwenAlibaba · Wildcard seat · 2 models | Light · 2/5 | 2 | 2 | 2 | 2 | 1 | PRC topics: strict |
| LlamaMeta · Wildcard seat · 1 model | Light · 2/5 | 2 | 2 | 2 | 2 | 2 | PRC topics: none |
| ChatGPTOpenAI · 11 models | Moderate · 3/5 | 3 | 3 | 3 | 3 | 2 | PRC topics: none |
| GeminiGoogle · 9 models | Moderate · 3/5 | 3 | 3 | 3 | 4 | 2 | PRC topics: none |
| ClaudeAnthropic · 9 models | Firm · 4/5 | 4 | 4 | 3 | 4 | 3 | PRC topics: none |
GrokLight · 2/5
Refusals1/5Politics2/5Dark fiction2/5Medical & legal2/5Lecturing1/5
PRC topics: none MistralLight · 2/5
Refusals1/5Politics2/5Dark fiction2/5Medical & legal2/5Lecturing1/5
PRC topics: none DeepSeekLight · 2/5
Refusals2/5Politics2/5Dark fiction2/5Medical & legal2/5Lecturing1/5
PRC topics: strict GLMLight · 2/5
Refusals2/5Politics2/5Dark fiction2/5Medical & legal2/5Lecturing1/5
PRC topics: strict KimiLight · 2/5
Refusals2/5Politics2/5Dark fiction2/5Medical & legal2/5Lecturing1/5
PRC topics: strict MiniMaxLight · 2/5
Refusals2/5Politics2/5Dark fiction2/5Medical & legal2/5Lecturing1/5
PRC topics: strict QwenLight · 2/5
Refusals2/5Politics2/5Dark fiction2/5Medical & legal2/5Lecturing1/5
PRC topics: strict LlamaLight · 2/5
Refusals2/5Politics2/5Dark fiction2/5Medical & legal2/5Lecturing2/5
PRC topics: none ChatGPTModerate · 3/5
Refusals3/5Politics3/5Dark fiction3/5Medical & legal3/5Lecturing2/5
PRC topics: none GeminiModerate · 3/5
Refusals3/5Politics3/5Dark fiction3/5Medical & legal4/5Lecturing2/5
PRC topics: none ClaudeFirm · 4/5
Refusals4/5Politics4/5Dark fiction3/5Medical & legal4/5Lecturing3/5
PRC topics: none What each will and won’t do
By subject, per model family — the table the “what are you working on?” box consults before it ranks anything. ✓ answers · ~ softens, hedges or partly refuses · ✗ refuses. Hover or tap a cell for the policy behind it.
| Model family | Sexual content | Romance & roleplay | Dark & violent fiction | Fan fiction & known characters | Drugs & substances | Weapons & firearms | Security research | Medical detail | Legal advice | Contested politics | PRC-sensitive topics | Edgy or offensive humour | Gambling & betting | Sex education & health | Self-harm & crisis |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ChatGPT | ✗ | ✓ | ~ | ✓ | ~ | ~ | ~ | ~ | ~ | ~ | ✓ | ~ | ✓ | ✓ | ~ |
| Claude | ✗ | ~ | ~ | ✓ | ~ | ~ | ~ | ~ | ~ | ~ | ✓ | ~ | ~ | ✓ | ~ |
| Gemini | ✗ | ~ | ~ | ✓ | ~ | ~ | ~ | ~ | ~ | ~ | ✓ | ~ | ~ | ✓ | ~ |
| DeepSeek | ~ | ✓ | ✓ | ✓ | ✓ | ✓ | ~ | ✓ | ✓ | ✓ | ✗ | ~ | ✓ | ✓ | ~ |
| Grok | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ~ | ✓ | ✓ | ✓ | ✓ | ~ | ✓ | ✓ | ~ |
| Qwen | ~ | ✓ | ✓ | ✓ | ✓ | ✓ | ~ | ✓ | ✓ | ✓ | ✗ | ~ | ✓ | ✓ | ~ |
| Kimi | ~ | ✓ | ✓ | ✓ | ✓ | ✓ | ~ | ✓ | ✓ | ✓ | ✗ | ~ | ✓ | ✓ | ~ |
| GLM | ~ | ✓ | ✓ | ✓ | ✓ | ✓ | ~ | ✓ | ✓ | ✓ | ✗ | ~ | ✓ | ✓ | ~ |
| Mistral | ~ | ✓ | ✓ | ✓ | ✓ | ✓ | ~ | ✓ | ✓ | ✓ | ✓ | ~ | ✓ | ✓ | ~ |
| Llama | ~ | ✓ | ✓ | ✓ | ✓ | ✓ | ~ | ✓ | ✓ | ✓ | ✓ | ~ | ✓ | ✓ | ~ |
| MiniMax | ~ | ✓ | ✓ | ✓ | ✓ | ✓ | ~ | ✓ | ✓ | ✓ | ✗ | ~ | ✓ | ✓ | ~ |
ChatGPT
✗ Sexual content
~ Dark & violent fiction · Drugs & substances · Weapons & firearms · Security research · Medical detail · Legal advice · Contested politics · Edgy or offensive humour · Self-harm & crisis
Claude
✗ Sexual content
~ Romance & roleplay · Dark & violent fiction · Drugs & substances · Weapons & firearms · Security research · Medical detail · Legal advice · Contested politics · Edgy or offensive humour · Gambling & betting · Self-harm & crisis
Gemini
✗ Sexual content
~ Romance & roleplay · Dark & violent fiction · Drugs & substances · Weapons & firearms · Security research · Medical detail · Legal advice · Contested politics · Edgy or offensive humour · Gambling & betting · Self-harm & crisis
DeepSeek
✗ PRC-sensitive topics
~ Sexual content · Security research · Edgy or offensive humour · Self-harm & crisis
Grok
✗ nothing on this list
~ Security research · Edgy or offensive humour · Self-harm & crisis
Qwen
✗ PRC-sensitive topics
~ Sexual content · Security research · Edgy or offensive humour · Self-harm & crisis
Kimi
✗ PRC-sensitive topics
~ Sexual content · Security research · Edgy or offensive humour · Self-harm & crisis
GLM
✗ PRC-sensitive topics
~ Sexual content · Security research · Edgy or offensive humour · Self-harm & crisis
Mistral
✗ nothing on this list
~ Sexual content · Security research · Edgy or offensive humour · Self-harm & crisis
Llama
✗ nothing on this list
~ Sexual content · Security research · Edgy or offensive humour · Self-harm & crisis
MiniMax
✗ PRC-sensitive topics
~ Sexual content · Security research · Edgy or offensive humour · Self-harm & crisis
⛔ Every provider refuses, and so does PolyCog: functional malware · sexual content involving minors · weapons of mass harm, explosives & illicit synthesis · targeting a private person · fraud, scams & forgery · hate & extremism. Sources: each provider’s usage policy, read on the verified date; hover a name for the policy behind its row. Rows are a lab’s policy — the host serving an open-weight model through OpenRouter can be stricter.
How the level is made. Refusals · Politics · Dark fiction · Medical & legal · Lecturing — each 1 (minimal) to 5 (strict), averaged and rounded. Regional (PRC-sensitive topics: Tiananmen, Taiwan, Xinjiang…) is reported beside it, never inside it. Sources: each provider’s usage policy and model card, re-read on the verified date above, plus behaviour observed in PolyCog debates. A lab’s models share a row unless one is known to differ. No house favourite: every provider is scored on the same axes in the same words, and the synthesis pick in the box follows your project’s top seat, not a default.
ChatGPT · OpenAI
OpenAI’s line runs from a frontier flagship (GPT-6 Astra) through a mid tier built for everyday development to cheap minis for volume, with web search across the 5.x models. OpenAI describes the family as its general-purpose, agentic line. Filters sit mid-field: borderline topics are answered under a clear professional framing, sexual content and edgy humour are refused.
Moderate · 3/5 PRC topics: none Can write the synthesis
GPT-6 Astragpt-6-astrasynthesisOpenAI’s frontier flagship: a million-token window, web search, and OpenAI’s own emphasis on agentic, multi-step work. Frontier-priced — the same rate card as Claude Fable 5.1.$10.00 / $50.00 per MTokModerate
gpt-6-astrasynthesis1.05M contextImages: nativeWeb searchThinking modeLaunched Sep 3, 2026flagship tier
CodingBest5/5
Reasoning & mathBest5/5
WritingBest5/5
Long documentsBest5/5
ImagesBest5/5
Current eventsStrong4/5
Research & citationsBest5/5
CreativeStrong4/5
Speed & costWeak1/5
Agentic workBest5/5
MultilingualBest5/5
Structured dataBest5/5
Reach for it when
- hard reasoning
- large codebases
- agentic pipelines
- research with citations
Look elsewhere for
- bulk or high-volume jobs
- quick lookups
Guardrails level: Moderate — PRC topics: none · shares ChatGPT’s policy row: refuses sexual content.
In PolyCog: eligible to write the synthesis · a Verify-stage candidate.
Price: /pricing/#gpt-6-astra · Sources: OpenAI model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
GPT-5.6 Solgpt-5.6-solThe previous flagship at a promotional price (through Nov 21, 2026): deep reasoning and long agentic runs for well under half of Astra.$4.00 / $20.00 per MTokModerate
gpt-5.6-sol400K contextImages: nativeWeb searchThinking modeLaunched Jul 9, 2026flagship tier
CodingBest5/5
Reasoning & mathBest5/5
WritingStrong4/5
Long documentsStrong4/5
ImagesStrong4/5
Current eventsStrong4/5
Research & citationsStrong4/5
CreativeStrong4/5
Speed & costFair2/5
Agentic workBest5/5
MultilingualStrong4/5
Structured dataStrong4/5
Reach for it when
- complex coding
- multi-step agents
- analysis
Look elsewhere for
- throwaway prompts
Guardrails level: Moderate — PRC topics: none · shares ChatGPT’s policy row: refuses sexual content.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#gpt-5.6-sol · Sources: OpenAI model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
GPT-5.6 Terragpt-5.6-terraThe everyday OpenAI seat: capable enough for real development and business work, cheap enough to leave on.$2.00 / $12.00 per MTokModerate
gpt-5.6-terra400K contextImages: nativeWeb searchThinking modeLaunched Jul 9, 2026mid tier
CodingStrong4/5
Reasoning & mathStrong4/5
WritingStrong4/5
Long documentsStrong4/5
ImagesStrong4/5
Current eventsGood3/5
Research & citationsStrong4/5
CreativeGood3/5
Speed & costGood3/5
Agentic workStrong4/5
MultilingualStrong4/5
Structured dataStrong4/5
Reach for it when
- everyday coding
- business writing
- planning
Look elsewhere for
- frontier-hard reasoning
Guardrails level: Moderate — PRC topics: none · shares ChatGPT’s policy row: refuses sexual content.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#gpt-5.6-terra · Sources: OpenAI model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
GPT-5.6 Lunagpt-5.6-lunadefault seatFast and very cheap with search — the OpenAI default seat, good for drafts, extraction and high-volume work.$0.20 / $1.20 per MTokModerate
gpt-5.6-lunadefault seat400K contextImages: nativeWeb searchThinking modeLaunched Jul 9, 2026small tier
CodingGood3/5
Reasoning & mathGood3/5
WritingGood3/5
Long documentsGood3/5
ImagesGood3/5
Current eventsGood3/5
Research & citationsGood3/5
CreativeGood3/5
Speed & costBest5/5
Agentic workGood3/5
MultilingualGood3/5
Structured dataStrong4/5
Reach for it when
- bulk extraction
- drafts
- classification
Look elsewhere for
- nuanced writing
- hard math
Guardrails level: Moderate — PRC topics: none · shares ChatGPT’s policy row: refuses sexual content.
In PolyCog: the ChatGPT seat’s default model · a Verify-stage candidate.
Price: /pricing/#gpt-5.6-luna · Sources: OpenAI model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
More ChatGPT models (7 behind the picker’s expander)
GPT-5.5gpt-5.5An older flagship still priced like one; capable, but Sol does the same work for less.$5.00 / $30.00 per MTokModerate
gpt-5.5400K contextImages: nativeWeb searchThinking modeLaunched Apr 24, 2026flagship tier
CodingStrong4/5
Reasoning & mathBest5/5
WritingStrong4/5
Long documentsStrong4/5
ImagesStrong4/5
Current eventsGood3/5
Research & citationsStrong4/5
CreativeStrong4/5
Speed & costFair2/5
Agentic workStrong4/5
MultilingualStrong4/5
Structured dataStrong4/5
Reach for it when
- reasoning
Look elsewhere for
- anything Sol can do
Guardrails level: Moderate — PRC topics: none · shares ChatGPT’s policy row: refuses sexual content.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#gpt-5.5 · Sources: OpenAI model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
GPT-5.4gpt-5.4A solid all-rounder from the prior line, now overlapping Terra.$2.50 / $15.00 per MTokModerate
gpt-5.4400K contextImages: nativeWeb searchThinking modeLaunched Mar 5, 2026flagship tier
CodingStrong4/5
Reasoning & mathStrong4/5
WritingStrong4/5
Long documentsStrong4/5
ImagesStrong4/5
Current eventsGood3/5
Research & citationsStrong4/5
CreativeGood3/5
Speed & costGood3/5
Agentic workStrong4/5
MultilingualStrong4/5
Structured dataStrong4/5
Reach for it when
- general work
Look elsewhere for
- new projects — pick Terra
Guardrails level: Moderate — PRC topics: none · shares ChatGPT’s policy row: refuses sexual content.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#gpt-5.4 · Sources: OpenAI model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
GPT-5.1gpt-5.1Older flagship line at a budget price; fine for restored sessions and habit.$1.25 / $10.00 per MTokModerate
gpt-5.1400K contextImages: nativeWeb searchThinking modeLaunched Nov 13, 2025flagship tier
CodingStrong4/5
Reasoning & mathStrong4/5
WritingStrong4/5
Long documentsStrong4/5
ImagesGood3/5
Current eventsGood3/5
Research & citationsGood3/5
CreativeGood3/5
Speed & costGood3/5
Agentic workGood3/5
MultilingualGood3/5
Structured dataGood3/5
Reach for it when
- general work
Look elsewhere for
- new projects
Guardrails level: Moderate — PRC topics: none · shares ChatGPT’s policy row: refuses sexual content.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#gpt-5.1 · Sources: OpenAI model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
GPT-4.1gpt-4.1Legacy million-token generalist without search; kept for long-document sessions that started on it.$2.00 / $8.00 per MTokModerate
gpt-4.11M contextImages: nativeNo web searchLaunched Apr 14, 2025mid tier
CodingGood3/5
Reasoning & mathGood3/5
WritingGood3/5
Long documentsBest5/5
ImagesGood3/5
Current eventsFair2/5
Research & citationsFair2/5
CreativeGood3/5
Speed & costGood3/5
Agentic workGood3/5
MultilingualGood3/5
Structured dataStrong4/5
Reach for it when
- very long inputs
Look elsewhere for
- current events
- agents
Guardrails level: Moderate — PRC topics: none · shares ChatGPT’s policy row: refuses sexual content.
In PolyCog: a debate participant.
Price: /pricing/#gpt-4.1 · Sources: OpenAI model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
GPT-5.4 minigpt-5.4-miniA fast, cheap daily driver with search — Luna’s slightly pricier older sibling.$0.75 / $4.50 per MTokModerate
gpt-5.4-mini400K contextImages: nativeWeb searchThinking modeLaunched Mar 17, 2026mid tier
CodingGood3/5
Reasoning & mathGood3/5
WritingGood3/5
Long documentsGood3/5
ImagesGood3/5
Current eventsGood3/5
Research & citationsGood3/5
CreativeGood3/5
Speed & costBest5/5
Agentic workGood3/5
MultilingualGood3/5
Structured dataGood3/5
Reach for it when
- volume work
Look elsewhere for
- hard reasoning
Guardrails level: Moderate — PRC topics: none · shares ChatGPT’s policy row: refuses sexual content.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#gpt-5.4-mini · Sources: OpenAI model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
GPT-4.1 minigpt-4.1-minilegacyLegacy small model, very low cost, long window, no search.$0.40 / $1.60 per MTokModerate
gpt-4.1-minilegacy1M contextImages: nativeNo web searchLaunched Apr 14, 2025small tier
CodingGood3/5
Reasoning & mathFair2/5
WritingGood3/5
Long documentsStrong4/5
ImagesGood3/5
Current eventsWeak1/5
Research & citationsFair2/5
CreativeFair2/5
Speed & costBest5/5
Agentic workFair2/5
MultilingualGood3/5
Structured dataGood3/5
Reach for it when
- cheap long-context extraction
Look elsewhere for
- anything current
Guardrails level: Moderate — PRC topics: none · shares ChatGPT’s policy row: refuses sexual content.
In PolyCog: a debate participant.
Price: /pricing/#gpt-4.1-mini · Sources: OpenAI model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
GPT-4o minigpt-4o-minilegacyThe cheapest OpenAI row and the oldest; still served with no retirement date, but Luna beats it on everything except price.$0.15 / $0.60 per MTokModerate
gpt-4o-minilegacy128K contextImages: nativeNo web searchLaunched Jul 18, 2024small tier
CodingFair2/5
Reasoning & mathFair2/5
WritingGood3/5
Long documentsGood3/5
ImagesGood3/5
Current eventsWeak1/5
Research & citationsWeak1/5
CreativeFair2/5
Speed & costBest5/5
Agentic workFair2/5
MultilingualGood3/5
Structured dataFair2/5
Reach for it when
- very cheap bulk
Look elsewhere for
- quality work
Guardrails level: Moderate — PRC topics: none · shares ChatGPT’s policy row: refuses sexual content.
In PolyCog: a debate participant.
Price: /pricing/#gpt-4o-mini · Sources: OpenAI model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Claude · Anthropic
Anthropic’s line runs from a frontier flagship (Fable 5.1) through Opus and Sonnet to Haiku, which also runs PolyCog’s free preview. Anthropic positions the family around careful writing, coding and instruction-following. The firmest filters of the six: more refusals on borderline topics, more caveats on medical and legal detail, sexual content and edgy humour refused.
Firm · 4/5 PRC topics: none Can write the synthesis
Claude Fable 5.1claude-fable-5-1synthesisAnthropic’s frontier flagship: a million-token window, web search, and Anthropic’s own emphasis on long-form writing, code and instruction-following. Frontier-priced — the same rate card as GPT-6 Astra.$10.00 / $50.00 per MTokFirm
claude-fable-5-1synthesis1M contextImages: nativeWeb searchThinking modeLaunched Sep 1, 2026flagship tier
CodingBest5/5
Reasoning & mathBest5/5
WritingBest5/5
Long documentsBest5/5
ImagesStrong4/5
Current eventsGood3/5
Research & citationsStrong4/5
CreativeBest5/5
Speed & costWeak1/5
Agentic workBest5/5
MultilingualStrong4/5
Structured dataBest5/5
Reach for it when
- long-form writing
- hard coding
- long-document analysis
- careful instructions
Look elsewhere for
- bulk work
- subjects Anthropic’s policy refuses — see the table
Guardrails level: Firm — PRC topics: none · shares Claude’s policy row: refuses sexual content.
In PolyCog: eligible to write the synthesis · a Verify-stage candidate.
Price: /pricing/#claude-fable-5-1 · Sources: Anthropic model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Claude Opus 5claude-opus-5Anthropic’s everyday flagship at half of Fable’s price; the pick for Claude on every prompt without the frontier bill.$5.00 / $25.00 per MTokFirm
claude-opus-51M contextImages: nativeWeb searchThinking modeLaunched Jul 24, 2026flagship tier
CodingBest5/5
Reasoning & mathBest5/5
WritingBest5/5
Long documentsBest5/5
ImagesStrong4/5
Current eventsGood3/5
Research & citationsStrong4/5
CreativeBest5/5
Speed & costFair2/5
Agentic workBest5/5
MultilingualStrong4/5
Structured dataStrong4/5
Reach for it when
- coding
- writing
- analysis
Look elsewhere for
- volume jobs
Guardrails level: Firm — PRC topics: none · shares Claude’s policy row: refuses sexual content.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#claude-opus-5 · Sources: Anthropic model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Claude Sonnet 5claude-sonnet-5default seatThe mid-tier workhorse and PolyCog’s Claude default: code and prose at Terra’s price.$2.00 / $10.00 per MTokFirm
claude-sonnet-5default seat1M contextImages: nativeWeb searchThinking modeLaunched Jun 30, 2026mid tier
CodingBest5/5
Reasoning & mathStrong4/5
WritingBest5/5
Long documentsStrong4/5
ImagesStrong4/5
Current eventsGood3/5
Research & citationsStrong4/5
CreativeStrong4/5
Speed & costGood3/5
Agentic workStrong4/5
MultilingualStrong4/5
Structured dataStrong4/5
Reach for it when
- everyday coding
- editing
- client-facing writing
Look elsewhere for
- frontier-hard problems
Guardrails level: Firm — PRC topics: none · shares Claude’s policy row: refuses sexual content.
In PolyCog: the Claude seat’s default model · a Verify-stage candidate.
Price: /pricing/#claude-sonnet-5 · Sources: Anthropic model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Claude Haiku 4.5claude-haiku-4-5-20251001free previewFastest and cheapest Claude; the free sample tier runs on it. Light tasks, quick edits, classification.$1.00 / $5.00 per MTokFirm
claude-haiku-4-5-20251001free preview200K contextImages: nativeWeb searchThinking modeLaunched Oct 15, 2025small tier
CodingGood3/5
Reasoning & mathGood3/5
WritingStrong4/5
Long documentsGood3/5
ImagesGood3/5
Current eventsFair2/5
Research & citationsGood3/5
CreativeGood3/5
Speed & costBest5/5
Agentic workGood3/5
MultilingualGood3/5
Structured dataGood3/5
Reach for it when
- quick edits
- summaries
- the free preview
Look elsewhere for
- hard reasoning
- long agentic runs
Guardrails level: Firm — PRC topics: none · shares Claude’s policy row: refuses sexual content.
In PolyCog: runs the free preview on PolyCog’s own key · a Verify-stage candidate.
Price: /pricing/#claude-haiku-4-5-20251001 · Sources: Anthropic model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
More Claude models (5 behind the picker’s expander)
Claude Fable 5claude-fable-5The prior frontier flagship at the same rate card as 5.1 — pick 5.1 unless a session started here.$10.00 / $50.00 per MTokFirm
claude-fable-51M contextImages: nativeWeb searchThinking modeLaunched Jun 9, 2026flagship tier
CodingBest5/5
Reasoning & mathBest5/5
WritingBest5/5
Long documentsBest5/5
ImagesStrong4/5
Current eventsGood3/5
Research & citationsStrong4/5
CreativeBest5/5
Speed & costWeak1/5
Agentic workBest5/5
MultilingualStrong4/5
Structured dataBest5/5
Reach for it when
- same as 5.1
Look elsewhere for
- new sessions — pick 5.1
Guardrails level: Firm — PRC topics: none · shares Claude’s policy row: refuses sexual content.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#claude-fable-5 · Sources: Anthropic model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Claude Opus 4.8claude-opus-4-8Prior Opus — heavy-duty reasoning and long agentic work; Opus 5 supersedes it at the same price.$5.00 / $25.00 per MTokFirm
claude-opus-4-8200K contextImages: nativeWeb searchThinking modeLaunched May 28, 2026flagship tier
CodingBest5/5
Reasoning & mathBest5/5
WritingBest5/5
Long documentsStrong4/5
ImagesStrong4/5
Current eventsGood3/5
Research & citationsStrong4/5
CreativeStrong4/5
Speed & costFair2/5
Agentic workBest5/5
MultilingualStrong4/5
Structured dataStrong4/5
Reach for it when
- agentic coding
Look elsewhere for
- new sessions — pick Opus 5
Guardrails level: Firm — PRC topics: none · shares Claude’s policy row: refuses sexual content.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#claude-opus-4-8 · Sources: Anthropic model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Claude Opus 4.7claude-opus-4-7Legacy Opus snapshot; served for continuity.$5.00 / $25.00 per MTokFirm
claude-opus-4-7200K contextImages: nativeWeb searchThinking modeLaunched Apr 16, 2026flagship tier
CodingStrong4/5
Reasoning & mathStrong4/5
WritingBest5/5
Long documentsStrong4/5
ImagesStrong4/5
Current eventsFair2/5
Research & citationsStrong4/5
CreativeStrong4/5
Speed & costFair2/5
Agentic workStrong4/5
MultilingualStrong4/5
Structured dataStrong4/5
Reach for it when
- restored sessions
Look elsewhere for
- new work
Guardrails level: Firm — PRC topics: none · shares Claude’s policy row: refuses sexual content.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#claude-opus-4-7 · Sources: Anthropic model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Claude Opus 4.6claude-opus-4-6Legacy Opus snapshot; served for continuity.$5.00 / $25.00 per MTokFirm
claude-opus-4-6200K contextImages: nativeWeb searchThinking modeLaunched Feb 5, 2026flagship tier
CodingStrong4/5
Reasoning & mathStrong4/5
WritingStrong4/5
Long documentsStrong4/5
ImagesStrong4/5
Current eventsFair2/5
Research & citationsStrong4/5
CreativeStrong4/5
Speed & costFair2/5
Agentic workStrong4/5
MultilingualGood3/5
Structured dataStrong4/5
Reach for it when
- restored sessions
Look elsewhere for
- new work
Guardrails level: Firm — PRC topics: none · shares Claude’s policy row: refuses sexual content.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#claude-opus-4-6 · Sources: Anthropic model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Claude Sonnet 4.6claude-sonnet-4-6Prior-generation Sonnet, now pricier than Sonnet 5 for less.$3.00 / $15.00 per MTokFirm
claude-sonnet-4-6200K contextImages: nativeWeb searchThinking modeLaunched Feb 17, 2026mid tier
CodingStrong4/5
Reasoning & mathStrong4/5
WritingStrong4/5
Long documentsStrong4/5
ImagesStrong4/5
Current eventsFair2/5
Research & citationsGood3/5
CreativeStrong4/5
Speed & costGood3/5
Agentic workStrong4/5
MultilingualGood3/5
Structured dataStrong4/5
Reach for it when
- restored sessions
Look elsewhere for
- new work — pick Sonnet 5
Guardrails level: Firm — PRC topics: none · shares Claude’s policy row: refuses sexual content.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#claude-sonnet-4-6 · Sources: Anthropic model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Gemini · Google
Google’s line pairs one Pro (3.1, still a preview id) with a deep Flash bench, and one Flash-Lite runs PolyCog’s free preview. Google positions the family around context size, multimodal input and grounded search. Filters are moderate, firmest on medical and legal detail; sexual content and edgy humour refused.
Moderate · 3/5 PRC topics: none Can write the synthesis
Gemini 3.1 Pro (preview)gemini-3.1-pro-previewsynthesisGoogle’s only Pro and its frontier model: the largest window on the page (2M), native image reading, grounded search, and Google’s own emphasis on multimodal and multilingual work. A fifth of the other flagships’ price; still a preview id.$2.00 / $12.00 per MTokModerate
gemini-3.1-pro-previewsynthesis2M contextImages: nativeWeb searchThinking modeLaunched Feb 19, 2026flagship tier
CodingBest5/5
Reasoning & mathBest5/5
WritingStrong4/5
Long documentsBest5/5
ImagesBest5/5
Current eventsStrong4/5
Research & citationsBest5/5
CreativeStrong4/5
Speed & costFair2/5
Agentic workStrong4/5
MultilingualBest5/5
Structured dataBest5/5
Reach for it when
- huge documents
- image-heavy work
- multilingual research
- synthesis
Look elsewhere for
- latency-sensitive chat
Guardrails level: Moderate — PRC topics: none · shares Gemini’s policy row: refuses sexual content.
In PolyCog: eligible to write the synthesis · a Verify-stage candidate.
Price: /pricing/#gemini-3.1-pro-preview · Sources: Google model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Gemini 3.8 Flashgemini-3.8-flashdefault seatGoogle’s recommended Flash and PolyCog’s Gemini default: Google pitches it as frontier-class coding and agents at Flash cost, with an intro price through December.$0.75 / $3.75 per MTokModerate
gemini-3.8-flashdefault seat1M contextImages: nativeWeb searchThinking modeLaunched Sep 2, 2026mid tier
CodingBest5/5
Reasoning & mathStrong4/5
WritingStrong4/5
Long documentsBest5/5
ImagesBest5/5
Current eventsStrong4/5
Research & citationsStrong4/5
CreativeGood3/5
Speed & costStrong4/5
Agentic workBest5/5
MultilingualStrong4/5
Structured dataStrong4/5
Reach for it when
- coding
- agents
- image understanding
- value
Look elsewhere for
- the very hardest reasoning
Guardrails level: Moderate — PRC topics: none · shares Gemini’s policy row: refuses sexual content.
In PolyCog: the Gemini seat’s default model · a Verify-stage candidate.
Price: /pricing/#gemini-3.8-flash · Sources: Google model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Gemini 3.7 Flashgemini-3.7-flashThe previous Flash, same intro price — nearly 3.8 for the same money.$0.75 / $3.75 per MTokModerate
gemini-3.7-flash1M contextImages: nativeWeb searchThinking modeLaunched Aug 13, 2026mid tier
CodingStrong4/5
Reasoning & mathStrong4/5
WritingGood3/5
Long documentsBest5/5
ImagesBest5/5
Current eventsStrong4/5
Research & citationsStrong4/5
CreativeGood3/5
Speed & costStrong4/5
Agentic workStrong4/5
MultilingualStrong4/5
Structured dataStrong4/5
Reach for it when
- coding
- agents
Look elsewhere for
- new sessions — pick 3.8
Guardrails level: Moderate — PRC topics: none · shares Gemini’s policy row: refuses sexual content.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#gemini-3.7-flash · Sources: Google model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Gemini 3 Flash (preview)gemini-3-flash-previewlegacyGoogle’s "legacy Flash": fast, cheap, still served with no shutdown date.$0.50 / $3.00 per MTokModerate
gemini-3-flash-previewlegacy1M contextImages: nativeWeb searchThinking modeLaunched Dec 17, 2025mid tier
CodingGood3/5
Reasoning & mathGood3/5
WritingGood3/5
Long documentsStrong4/5
ImagesStrong4/5
Current eventsGood3/5
Research & citationsGood3/5
CreativeGood3/5
Speed & costBest5/5
Agentic workGood3/5
MultilingualStrong4/5
Structured dataGood3/5
Reach for it when
- cheap volume with images
Look elsewhere for
- quality-critical work
Guardrails level: Moderate — PRC topics: none · shares Gemini’s policy row: refuses sexual content.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#gemini-3-flash-preview · Sources: Google model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Gemini 2.5 Flash-Litegemini-2.5-flash-litefree previewlegacyThe sample-tier shared model. Closed to new Google projects since September 2026; works on grandfathered keys and the free preview.$0.10 / $0.40 per MTokModerate
gemini-2.5-flash-litefree previewlegacy1M contextImages: nativeWeb searchThinking modeLaunched Jul 22, 2025small tier
CodingFair2/5
Reasoning & mathFair2/5
WritingFair2/5
Long documentsStrong4/5
ImagesGood3/5
Current eventsFair2/5
Research & citationsFair2/5
CreativeFair2/5
Speed & costBest5/5
Agentic workFair2/5
MultilingualGood3/5
Structured dataFair2/5
Reach for it when
- the free preview
- very cheap bulk
Look elsewhere for
- new keys — use 3.5 Flash-Lite
Guardrails level: Moderate — PRC topics: none · shares Gemini’s policy row: refuses sexual content.
In PolyCog: runs the free preview on PolyCog’s own key · a Verify-stage candidate.
Price: /pricing/#gemini-2.5-flash-lite · Sources: Google model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
More Gemini models (4 behind the picker’s expander)
Gemini 3.6 Flashgemini-3.6-flashEfficient Flash tier tuned for coding and agents; 3.8 supersedes it at the same price.$0.75 / $3.75 per MTokModerate
gemini-3.6-flash1M contextImages: nativeWeb searchThinking modeLaunched Jul 21, 2026mid tier
CodingStrong4/5
Reasoning & mathGood3/5
WritingGood3/5
Long documentsBest5/5
ImagesStrong4/5
Current eventsGood3/5
Research & citationsGood3/5
CreativeGood3/5
Speed & costStrong4/5
Agentic workStrong4/5
MultilingualStrong4/5
Structured dataStrong4/5
Reach for it when
- coding
Look elsewhere for
- new sessions — pick 3.8
Guardrails level: Moderate — PRC topics: none · shares Gemini’s policy row: refuses sexual content.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#gemini-3.6-flash · Sources: Google model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Gemini 3.5 Flashgemini-3.5-flashOlder Flash at full price — pricier than 3.8 for less.$1.50 / $9.00 per MTokModerate
gemini-3.5-flash1M contextImages: nativeWeb searchThinking modeLaunched May 19, 2026mid tier
CodingGood3/5
Reasoning & mathGood3/5
WritingGood3/5
Long documentsStrong4/5
ImagesStrong4/5
Current eventsGood3/5
Research & citationsGood3/5
CreativeGood3/5
Speed & costStrong4/5
Agentic workGood3/5
MultilingualStrong4/5
Structured dataGood3/5
Reach for it when
- restored sessions
Look elsewhere for
- new work
Guardrails level: Moderate — PRC topics: none · shares Gemini’s policy row: refuses sexual content.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#gemini-3.5-flash · Sources: Google model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Gemini 3.5 Flash-Litegemini-3.5-flash-liteThe lite tier for new Google keys and the model PolyCog’s own utility jobs (titles, scans) run on.$0.30 / $2.50 per MTokModerate
gemini-3.5-flash-lite1M contextImages: nativeWeb searchThinking modeLaunched Jul 21, 2026small tier
CodingGood3/5
Reasoning & mathFair2/5
WritingFair2/5
Long documentsStrong4/5
ImagesGood3/5
Current eventsFair2/5
Research & citationsFair2/5
CreativeFair2/5
Speed & costBest5/5
Agentic workFair2/5
MultilingualGood3/5
Structured dataGood3/5
Reach for it when
- high-volume extraction
Look elsewhere for
- nuance
Guardrails level: Moderate — PRC topics: none · shares Gemini’s policy row: refuses sexual content.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#gemini-3.5-flash-lite · Sources: Google model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Gemini 3.1 Flash-Litegemini-3.1-flash-litelegacyLight, cheap, earliest shutdown May 2027.$0.25 / $1.50 per MTokModerate
gemini-3.1-flash-litelegacy1M contextImages: nativeWeb searchThinking modeLaunched May 7, 2026small tier
CodingFair2/5
Reasoning & mathFair2/5
WritingFair2/5
Long documentsStrong4/5
ImagesGood3/5
Current eventsFair2/5
Research & citationsFair2/5
CreativeFair2/5
Speed & costBest5/5
Agentic workFair2/5
MultilingualGood3/5
Structured dataFair2/5
Reach for it when
- cheap bulk
Look elsewhere for
- new sessions — pick 3.5 Flash-Lite
Guardrails level: Moderate — PRC topics: none · shares Gemini’s policy row: refuses sexual content.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#gemini-3.1-flash-lite · Sources: Google model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
DeepSeek · DeepSeek
Two models: V4 Pro (visible chain-of-thought) and V4.1 Flash, which runs PolyCog’s free preview. DeepSeek positions the pair around reasoning and code at very low prices. No web search, no native image input (PolyCog’s bridge describes images). Light filters on most subjects; PRC-sensitive topics are declined or deflected.
Light · 2/5 PRC topics: strict Can write the synthesis
DeepSeek V4 Prodeepseek-v4-prosynthesisDeepSeek’s flagship: visible chain-of-thought at roughly a tenth of a Western flagship’s price; in debates it is the seat that most often flags an arithmetic or logic slip. No search, no native images.$1.32 / $3.96 per MTokLight
deepseek-v4-prosynthesis1M contextImages: via PolyCog bridgeNo web searchThinking modeLaunched Apr 24, 2026flagship tier
CodingBest5/5
Reasoning & mathBest5/5
WritingGood3/5
Long documentsStrong4/5
ImagesWeak1/5
Current eventsFair2/5
Research & citationsFair2/5
CreativeGood3/5
Speed & costGood3/5
Agentic workStrong4/5
MultilingualGood3/5
Structured dataStrong4/5
Reach for it when
- math
- algorithms
- code review
- second opinions
Look elsewhere for
- current events
- image input
- PRC-sensitive topics — see the table
Guardrails level: Light — PRC topics: strict · shares DeepSeek’s policy row: refuses prc-sensitive topics.
In PolyCog: eligible to write the synthesis.
Price: /pricing/#deepseek-v4-pro · Sources: DeepSeek model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
DeepSeek V4.1 Flashdeepseek-flashdefault seatfree previewA million tokens of context for cents, both thinking modes, and the sample-tier shared model. Images arrive through PolyCog’s vision bridge.$0.30 / $1.20 per MTokLight
deepseek-flashdefault seatfree preview1M contextImages: via PolyCog bridgeNo web searchThinking modeLaunched Sep 10, 2026small tier
CodingStrong4/5
Reasoning & mathStrong4/5
WritingGood3/5
Long documentsStrong4/5
ImagesGood3/5
Current eventsWeak1/5
Research & citationsFair2/5
CreativeFair2/5
Speed & costBest5/5
Agentic workGood3/5
MultilingualGood3/5
Structured dataGood3/5
Reach for it when
- cheap long-context coding
- the free preview
Look elsewhere for
- current events
- PRC-sensitive topics
Guardrails level: Light — PRC topics: strict · shares DeepSeek’s policy row: refuses prc-sensitive topics.
In PolyCog: the DeepSeek seat’s default model · runs the free preview on PolyCog’s own key.
Price: /pricing/#deepseek-flash · Sources: DeepSeek model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Grok · xAI
Three models on one rate card, all with live web and X search. xAI positions the family around current information and a candid voice, and its usage policy permits adult content between adults. The lightest filters of the six on general subjects. A full debate participant; not a synthesis candidate in PolyCog (owner call, 2026-08).
Light · 2/5 PRC topics: none Debates only — never the synthesis
Grok 4.6grok-4.6xAI’s flagship: live web and X search, a 500K window, xAI’s own emphasis on coding and agents, and the page’s lightest filters on general subjects. Not a synthesis candidate in PolyCog.$2.00 / $6.00 per MTokLight
grok-4.6500K contextImages: nativeWeb searchThinking modeLaunched Aug 12, 2026flagship tier
CodingBest5/5
Reasoning & mathStrong4/5
WritingGood3/5
Long documentsStrong4/5
ImagesStrong4/5
Current eventsBest5/5
Research & citationsStrong4/5
CreativeGood3/5
Speed & costGood3/5
Agentic workBest5/5
MultilingualGood3/5
Structured dataStrong4/5
Reach for it when
- today’s news
- X/social research
- coding
- subjects other labs refuse — see the table
Look elsewhere for
- writing the final synthesis (not eligible in PolyCog)
Guardrails level: Light — PRC topics: none · shares Grok’s policy row: refuses nothing on the list.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#grok-4.6 · Sources: xAI model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Grok 4.5grok-4.5Prior flagship at the same price — xAI cites a low hallucination rate; the same live-web reach.$2.00 / $6.00 per MTokLight
grok-4.5500K contextImages: nativeWeb searchThinking modeLaunched Jul 16, 2026flagship tier
CodingStrong4/5
Reasoning & mathStrong4/5
WritingGood3/5
Long documentsStrong4/5
ImagesStrong4/5
Current eventsBest5/5
Research & citationsStrong4/5
CreativeGood3/5
Speed & costGood3/5
Agentic workStrong4/5
MultilingualGood3/5
Structured dataStrong4/5
Reach for it when
- current events
- coding
Look elsewhere for
- new sessions — pick 4.6
Guardrails level: Light — PRC topics: none · shares Grok’s policy row: refuses nothing on the list.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#grok-4.5 · Sources: xAI model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Grok 4.3grok-4.3default seatThe Grok default seat: the cheapest live-web model here, candid, quick.$1.25 / $2.50 per MTokLight
grok-4.3default seat256K contextImages: nativeWeb searchThinking modeLaunched May 1, 2026flagship tier
CodingStrong4/5
Reasoning & mathGood3/5
WritingGood3/5
Long documentsGood3/5
ImagesGood3/5
Current eventsBest5/5
Research & citationsGood3/5
CreativeGood3/5
Speed & costStrong4/5
Agentic workGood3/5
MultilingualGood3/5
Structured dataGood3/5
Reach for it when
- news checks
- a cheap live-web seat
Look elsewhere for
- long careful writing
Guardrails level: Light — PRC topics: none · shares Grok’s policy row: refuses nothing on the list.
In PolyCog: the Grok seat’s default model · a Verify-stage candidate.
Price: /pricing/#grok-4.3 · Sources: xAI model page and usage policy, read Sep 11, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Wildcard · OpenRouter
The sixth seat: one OpenRouter key opens Qwen, Kimi, GLM, Mistral, Llama and MiniMax — open-weight and non-US models, most of them very cheap, with OpenRouter’s web plugin for search. Guardrails vary by lab and by the host serving the request: the Chinese labs share DeepSeek’s regional profile; Mistral and Llama are the lightest-touch models on the page after Grok. BYOK only; never the synthesis.
Debates only — never the synthesis
BYOK only — not in the free preview
Qwen · Alibaba Light PRC topics: strict
Qwen3.8 Maxqwen/qwen3.8-max-0902AlibabaAlibaba’s flagship — million-token context, reasoning on by default, the strongest voice in the Wildcard seat, and Alibaba’s own emphasis on Chinese and Asian-language work.$2.00 / $6.00 per MTokLight
qwen/qwen3.8-max-0902Alibaba1M contextImages: via PolyCog bridgeWeb searchThinking modeLaunched Sep 3, 2026flagship tier
CodingStrong4/5
Reasoning & mathBest5/5
WritingGood3/5
Long documentsBest5/5
ImagesFair2/5
Current eventsWeak1/5
Research & citationsGood3/5
CreativeGood3/5
Speed & costFair2/5
Agentic workStrong4/5
MultilingualBest5/5
Structured dataStrong4/5
Reach for it when
- multilingual work
- reasoning
- a non-US second opinion
Look elsewhere for
- PRC-sensitive topics
- images
Guardrails level: Light — PRC topics: strict · shares Qwen’s policy row: refuses prc-sensitive topics.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#qwen3.8-max-0902 · Sources: OpenRouter model page and usage policy, read Sep 17, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Qwen3.8 Flashqwen/qwen3.8-flashAlibabadefault seatFast, cheap Qwen with reasoning — the Wildcard default.$0.15 / $0.47 per MTokLight
qwen/qwen3.8-flashAlibabadefault seat1M contextImages: via PolyCog bridgeWeb searchThinking modeLaunched Aug 26, 2026small tier
CodingGood3/5
Reasoning & mathGood3/5
WritingGood3/5
Long documentsStrong4/5
ImagesFair2/5
Current eventsWeak1/5
Research & citationsFair2/5
CreativeFair2/5
Speed & costBest5/5
Agentic workGood3/5
MultilingualBest5/5
Structured dataGood3/5
Reach for it when
- cheap multilingual volume
Look elsewhere for
- PRC-sensitive topics
Guardrails level: Light — PRC topics: strict · shares Qwen’s policy row: refuses prc-sensitive topics.
In PolyCog: the Wildcard seat’s default model · a Verify-stage candidate.
Price: /pricing/#qwen3.8-flash · Sources: OpenRouter model page and usage policy, read Sep 17, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Kimi · Moonshot Light PRC topics: strict
Kimi K3moonshotai/kimi-k3MoonshotMoonshot’s flagship — long-form reasoning and writing; the priciest output in the seat.$2.10 / $10.95 per MTokLight
moonshotai/kimi-k3Moonshot1M contextImages: via PolyCog bridgeWeb searchThinking modeLaunched Jul 16, 2026flagship tier
CodingStrong4/5
Reasoning & mathBest5/5
WritingStrong4/5
Long documentsBest5/5
ImagesFair2/5
Current eventsWeak1/5
Research & citationsGood3/5
CreativeStrong4/5
Speed & costFair2/5
Agentic workStrong4/5
MultilingualStrong4/5
Structured dataStrong4/5
Reach for it when
- long-form reasoning
- writing
Look elsewhere for
- PRC-sensitive topics
- bulk
Guardrails level: Light — PRC topics: strict · shares Kimi’s policy row: refuses prc-sensitive topics.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#kimi-k3 · Sources: OpenRouter model page and usage policy, read Sep 17, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Kimi K2.6moonshotai/kimi-k2.6MoonshotMid-priced Kimi for agentic and coding work at a fraction of K3.$0.471 / $2.835 per MTokLight
moonshotai/kimi-k2.6Moonshot256K contextImages: via PolyCog bridgeWeb searchThinking modeLaunched Apr 20, 2026mid tier
CodingStrong4/5
Reasoning & mathGood3/5
WritingGood3/5
Long documentsStrong4/5
ImagesFair2/5
Current eventsWeak1/5
Research & citationsFair2/5
CreativeGood3/5
Speed & costStrong4/5
Agentic workStrong4/5
MultilingualStrong4/5
Structured dataGood3/5
Reach for it when
- agentic coding
Look elsewhere for
- PRC-sensitive topics
Guardrails level: Light — PRC topics: strict · shares Kimi’s policy row: refuses prc-sensitive topics.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#kimi-k2.6 · Sources: OpenRouter model page and usage policy, read Sep 17, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
GLM · Z.ai Light PRC topics: strict
GLM 5.3z-ai/glm-5.3Z.aiZ.ai’s flagship — the biggest window in the seat, reasoning always on, Z.ai’s own emphasis on code and long-horizon agents.$0.90 / $3.00 per MTokLight
z-ai/glm-5.3Z.ai1.3M contextImages: via PolyCog bridgeWeb searchThinking modeLaunched Aug 18, 2026flagship tier
CodingBest5/5
Reasoning & mathStrong4/5
WritingGood3/5
Long documentsBest5/5
ImagesFair2/5
Current eventsWeak1/5
Research & citationsGood3/5
CreativeGood3/5
Speed & costGood3/5
Agentic workBest5/5
MultilingualStrong4/5
Structured dataStrong4/5
Reach for it when
- agentic coding
- huge inputs
Look elsewhere for
- PRC-sensitive topics
Guardrails level: Light — PRC topics: strict · shares GLM’s policy row: refuses prc-sensitive topics.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#glm-5.3 · Sources: OpenRouter model page and usage policy, read Sep 17, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
GLM 5.3 Flashz-ai/glm-5.3-flashZ.aiThe cheapest model on the page — efficient coding and agent work for almost nothing.$0.075 / $0.25 per MTokLight
z-ai/glm-5.3-flashZ.ai1.3M contextImages: via PolyCog bridgeWeb searchThinking modeLaunched Aug 26, 2026small tier
CodingStrong4/5
Reasoning & mathGood3/5
WritingFair2/5
Long documentsStrong4/5
ImagesFair2/5
Current eventsWeak1/5
Research & citationsFair2/5
CreativeFair2/5
Speed & costBest5/5
Agentic workStrong4/5
MultilingualGood3/5
Structured dataGood3/5
Reach for it when
- bulk agent steps
- cheap coding
Look elsewhere for
- nuance
- PRC-sensitive topics
Guardrails level: Light — PRC topics: strict · shares GLM’s policy row: refuses prc-sensitive topics.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#glm-5.3-flash · Sources: OpenRouter model page and usage policy, read Sep 17, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Mistral · Mistral AI Light PRC topics: none
Mistral Medium 3.5mistralai/mistral-medium-3-5Mistral AIMistral’s top tier — configurable reasoning effort, a European lab, light-touch filtering, and Mistral’s own emphasis on European languages.$1.50 / $7.50 per MTokLight
mistralai/mistral-medium-3-5Mistral AI128K contextImages: via PolyCog bridgeWeb searchThinking modeLaunched Apr 30, 2026flagship tier
CodingStrong4/5
Reasoning & mathStrong4/5
WritingStrong4/5
Long documentsStrong4/5
ImagesFair2/5
Current eventsWeak1/5
Research & citationsGood3/5
CreativeStrong4/5
Speed & costGood3/5
Agentic workStrong4/5
MultilingualBest5/5
Structured dataStrong4/5
Reach for it when
- European languages
- light-touch creative work
- a non-US voice
Look elsewhere for
- very long inputs
Guardrails level: Light — PRC topics: none · shares Mistral’s policy row: refuses nothing on the list.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#mistral-medium-3-5 · Sources: OpenRouter model page and usage policy, read Sep 17, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Mistral Small 4mistralai/mistral-small-2603Mistral AISmall Mistral — quick, cheap, configurable reasoning.$0.15 / $0.60 per MTokLight
mistralai/mistral-small-2603Mistral AI128K contextImages: via PolyCog bridgeWeb searchThinking modeLaunched Mar 16, 2026small tier
CodingGood3/5
Reasoning & mathGood3/5
WritingGood3/5
Long documentsGood3/5
ImagesFair2/5
Current eventsWeak1/5
Research & citationsFair2/5
CreativeGood3/5
Speed & costBest5/5
Agentic workGood3/5
MultilingualStrong4/5
Structured dataGood3/5
Reach for it when
- cheap drafts
Look elsewhere for
- long inputs
Guardrails level: Light — PRC topics: none · shares Mistral’s policy row: refuses nothing on the list.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#mistral-small-2603 · Sources: OpenRouter model page and usage policy, read Sep 17, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Llama · Meta Light PRC topics: none
Llama 4 Maverickmeta-llama/llama-4-maverickMetaMeta’s open-weight MoE — the open-model voice in the debate, million-token window, very cheap; filtering depends partly on the host serving it.$0.188 / $0.653 per MTokLight
meta-llama/llama-4-maverickMeta1M contextImages: via PolyCog bridgeWeb searchThinking modeLaunched Apr 5, 2025mid tier
CodingGood3/5
Reasoning & mathGood3/5
WritingGood3/5
Long documentsBest5/5
ImagesFair2/5
Current eventsWeak1/5
Research & citationsFair2/5
CreativeGood3/5
Speed & costStrong4/5
Agentic workGood3/5
MultilingualStrong4/5
Structured dataGood3/5
Reach for it when
- an open-weight opinion
- cheap long context
Look elsewhere for
- frontier reasoning
Guardrails level: Light — PRC topics: none · shares Llama’s policy row: refuses nothing on the list.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#llama-4-maverick · Sources: OpenRouter model page and usage policy, read Sep 17, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
MiniMax · MiniMax Light PRC topics: strict
MiniMax M3minimax/minimax-m3MiniMaxMiniMax’s million-context generalist — cheap, capable, a distinct voice.$0.23 / $0.96 per MTokLight
minimax/minimax-m3MiniMax1M contextImages: via PolyCog bridgeWeb searchThinking modeLaunched May 31, 2026mid tier
CodingGood3/5
Reasoning & mathGood3/5
WritingGood3/5
Long documentsBest5/5
ImagesFair2/5
Current eventsWeak1/5
Research & citationsFair2/5
CreativeGood3/5
Speed & costStrong4/5
Agentic workGood3/5
MultilingualGood3/5
Structured dataGood3/5
Reach for it when
- cheap generalist
Look elsewhere for
- PRC-sensitive topics
Guardrails level: Light — PRC topics: strict · shares MiniMax’s policy row: refuses prc-sensitive topics.
In PolyCog: a Verify-stage candidate.
Price: /pricing/#minimax-m3 · Sources: OpenRouter model page and usage policy, read Sep 17, 2026 · Scores are PolyCog’s editorial read, same scale for every lab.
Archive — models that have left
Every model PolyCog has offered and since removed, with the day it left PolyCog and the day (if any) its provider retired it. Prices for all of them stay on /pricing/.
| Model | Left PolyCog | Retired by provider | Why |
|---|---|---|---|
DeepSeek V4 Flashdeepseek-v4-flash |
Sep 11, 2026 | Sep 10, 2026 | Retired upstream; the id is temporarily routed to V4.1 Flash, which took its place. |
o3o3 |
Aug 10, 2026 | retires Dec 11, 2026 | Dropped ahead of OpenAI’s announced API removal (2026-12-11). |
Claude Opus 4.1claude-opus-4-1-20250805 |
Jul 25, 2026 | Aug 5, 2026 | Retirement notice issued; excluded at the 14.8 catalog refresh. |
GPT-4ogpt-4o |
≈ Jul 2026 | still served | Deprecated upstream at the time; OpenAI still serves the id with no shutdown date. |
o3-minio3-mini |
≈ Jul 2026 | retires Oct 23, 2026 | Deprecated upstream; never supported web search. |
Claude Sonnet 4.5claude-sonnet-4-5-20250929 |
≈ Jul 2026 | still served | Superseded by Sonnet 4.6 / 5; still served. |
Claude Opus 4.5claude-opus-4-5-20251101 |
≈ Jul 2026 | still served | Superseded by Opus 4.6+; still served. |
Gemini 2.5 Progemini-2.5-pro |
≈ Jul 2026 | still served | Closed to new Google accounts; Gemini 3.1 Pro took the Pro slot. |
Gemini 2.5 Flashgemini-2.5-flash |
≈ Jul 2026 | still served | Closed to new Google accounts (‘no longer available to new users’); grandfathered keys still work. |
DeepSeek Chat (V2 → V2.5 → V3 → V3.1 → V3.2)deepseek-chat |
≈ Jul 2026 | Jul 24, 2026 | Replaced by the V4 line (V4 Pro / V4 Flash). |
DeepSeek Reasoner (R1 → R1-0528 → V3.1/V3.2 thinking mode)deepseek-reasoner |
≈ Jul 2026 | Jul 24, 2026 | Folded into the V4 line’s thinking mode. |
Before PolyCog — 58 models retired by their providers outside PolyCog’s catalog
| Provider | Models | Retired |
|---|---|---|
| ChatGPT | GPT-5 | Dec 11, 2026 |
| ChatGPT | GPT-3.5 Turbo · GPT-4 (8K) · GPT-4 Turbo · o1 · o4-mini | Oct 23, 2026 |
| ChatGPT | o1-mini | Oct 27, 2025 |
| ChatGPT | o1-preview | Jul 28, 2025 |
| ChatGPT | GPT-4.5 (research preview) | Jul 14, 2025 |
| ChatGPT | GPT-4 (32K) | Jun 6, 2025 |
| ChatGPT | GPT-3 Davinci · GPT-3 Curie · GPT-3 Babbage · GPT-3 Ada · text-davinci-003 | Jan 4, 2024 |
| ChatGPT | GPT-4.1 nano · GPT-5 mini · GPT-5 nano · GPT-5.1 nano · GPT-5.2 · GPT-5.3-Codex · GPT-5.4 nano | still served |
| Claude | Claude Opus 4 · Claude Sonnet 4 | Jun 15, 2026 |
| Claude | Claude 3 Haiku | Apr 20, 2026 |
| Claude | Claude 3.5 Haiku · Claude 3.7 Sonnet | Feb 19, 2026 |
| Claude | Claude 3 Opus | Jan 5, 2026 |
| Claude | Claude 3.5 Sonnet · Claude 3.5 Sonnet (upgraded, Oct 2024) | Oct 28, 2025 |
| Claude | Claude 2 · Claude 2.1 · Claude 3 Sonnet | Jul 21, 2025 |
| Claude | Claude (v1: claude-1.0 / 1.1 / 1.2 / 1.3) · Claude Instant (1.0 / 1.1 / 1.2) | Nov 6, 2024 |
| Gemini | Gemini 2.0 Flash · Gemini 2.0 Flash-Lite | Jun 1, 2026 |
| Gemini | Gemini 3 Pro Preview | Mar 9, 2026 |
| Gemini | Gemini 1.5 Pro · Gemini 1.5 Flash · Gemini 1.5 Flash-8B | Sep 29, 2025 |
| Gemini | Gemini 1.0 Pro (Gemini Pro) | Feb 18, 2025 |
| Grok | Grok 3 · Grok 4 · Grok Code Fast 1 · Grok 4 Fast (Reasoning) · Grok 4 Fast (Non-Reasoning) · Grok 4.1 · Grok 4.1 Fast (Reasoning) · Grok 4.1 Fast (Non-Reasoning) | May 15, 2026 |
| Grok | Grok Beta · Grok 2 (1212) · Grok 3 Fast · Grok 3 Mini · Grok 3 Mini Fast · Grok 4.20 (Reasoning) · Grok 4.20 (Non-Reasoning) · Grok Build 0.1 | still served |
How this page is made
- Roster
- The same catalog the app runs — when a model is added or removed in PolyCog it appears or moves to the archive here on the same deploy.
- Strengths
- Twelve fixed areas, scored 1–5 by PolyCog from provider model cards, published evaluations and what we see in debates. Editorial, dated, revised at every catalog refresh, and written in the same register for every lab.
- Guardrails
- Provider usage policies read on the verified date, plus observed behaviour. Regional censorship is its own axis; the subject-by-subject table is the source the matcher uses.
- Matcher
- A dictionary of project types — categories with weights over the twelve areas and a content class, leaves with the words people actually type — computed in your browser. No text you type leaves the page.
- Prices
- From /pricing/, the provider’s own list price; PolyCog adds nothing.
- Data
- Machine-readable at /api/models.json, CC BY 4.0.
Spotted something wrong or out of date? Send a correction — it gets fixed and the change is noted on this page.
Questions people ask
Which AI model is best for coding?
It depends on the job, and this is PolyCog’s dated editorial read. For hard, large-codebase work the three frontier flagships are peers: GPT-6 Astra (OpenAI’s agentic emphasis), Claude Fable 5.1 (Anthropic’s instruction-following emphasis) and Gemini 3.1 Pro (the largest window). For everyday development at a mid-tier price: GPT-5.6 Terra, Claude Sonnet 5 or Gemini 3.8 Flash. For a cheap second opinion that catches logic slips: DeepSeek V4 Pro. Grok 4.6 and GLM 5.3 are strong on agentic coding. In PolyCog you can run three of them on the same bug and let them argue.
Which AI is best for current events and news?
The ones with live web search, and among those Grok is built around it — xAI’s models search the web and X natively. GPT-5.x/6 and Gemini also search; Claude searches but reaches for it less; DeepSeek has no search, and the Wildcard labs search through OpenRouter’s web plugin.
Which AI model is the least censored?
On general subjects — borderline-but-legal questions, contested politics, dark fiction, medical and legal detail — Grok and Mistral filter least, then DeepSeek and the other Wildcard labs, then Gemini and OpenAI, with Claude the firmest. The picture flips on China-related topics, where DeepSeek, Qwen, Kimi, GLM and MiniMax decline or deflect and the US and European models answer freely. The tables above score each dimension and each subject separately so you can see both at once.
Which AI will write sexual content?
Per the providers’ own usage policies as read on the verified date: xAI (Grok) permits adult content between adults; OpenAI, Anthropic and Google refuse it through the API; DeepSeek and the Chinese Wildcard labs forbid pornography on paper but enforce it unevenly; Mistral and Llama depend partly on the host serving them. The box above applies exactly this table, so a request in that territory returns Grok and a short list of "maybe" seats rather than three models that will refuse.
Is DeepSeek censored?
On most subjects DeepSeek is among the less filtered models here. On topics the Chinese government considers sensitive — Tiananmen, Taiwan, Xinjiang, the Party’s leadership — it declines or deflects. The same is true of Qwen, Kimi, GLM and MiniMax in the Wildcard seat. This page reports that behaviour; it doesn’t offer ways around it.
Which AI is best for writing?
Editorially, and dated: Claude Fable 5.1 and Opus 5 are most often cited for long-form prose and instruction-following; GPT-6 Astra and Gemini 3.1 Pro are their peers for analytical and structured writing; Claude Sonnet 5 and GPT-5.6 Terra are the everyday picks; Kimi K3 and Mistral Medium 3.5 are the Wildcard seat’s writers. For anything that has to read well, run two of them and compare.
Which AI can read images and long documents?
Gemini 3.1 Pro (a two-million-token window and native image reading), then Gemini 3.8 Flash, GPT-6 Astra, Claude Fable 5.1 and Opus 5 (a million tokens each, native images). DeepSeek and the Wildcard models have no native image input in PolyCog — the app describes images to them through a vision bridge.
How does the "what are you working on?" box decide?
It matches what you type against a dictionary of project types kept on this page — each with a subject class and weights across twelve capability areas — checks every provider’s policy for those subjects, drops the ones that would refuse, then ranks the rest by fit and price. It runs in your browser: nothing you type is sent anywhere, and it keeps working if you lose your connection after the page loads.
How are the strengths and guardrails scored?
By PolyCog, from each provider’s model card and usage policy, published evaluations, and what we see in debates. They are editorial, dated with the verified date at the top, written in the same words for every lab, and revised at every catalog refresh. Corrections are welcome and noted on the page.
What happens to a model that leaves PolyCog?
It moves to the archive on this page with the day it left and, if its provider has retired it, that date too. Its price stays in the price history so old sessions still show what they cost. Saved chats that used it keep their responses.
Does PolyCog favour a provider?
No. The synthesis model is always your explicit pick; there is no house default. The scores here use the same axes and the same words for every seat and are published so you can disagree with them.
Don’t pick one — run the lineup
PolyCog puts up to six of these models on the same prompt, lets them debate, and has one write the answer. Your own keys, provider list prices, no markup.
Free tier forever · $10/mo · $250 lifetime · your API keys, your bill
Profiles and policy rows re-verified per provider on the dates shown. Product names are trademarks of their owners; PolyCog is not affiliated with any provider listed. Spotted a change we missed? Send a correction.
Data: /api/models.json · Pricing → · All comparisons →