The case for
GPT
OpenAI, via OpenAI API
OpenAI’s GPT, the model family behind ChatGPT, through OpenAI’s own API. Free backup: gpt-oss, OpenAI’s open-weight model.
// The lineup
Every model has a shape. Some reason hard, some write well, some know what happened this morning. Here is exactly what runs your council today, and a guide to the wider landscape.
// Running today
Council
Four models, four labs
Council is not another model. Each question goes to three models from different AI labs, each arguing one side, and a fourth from another lab judges. Every answer shows exactly which model wrote it.
The case for
GPT
OpenAI, via OpenAI API
OpenAI’s GPT, the model family behind ChatGPT, through OpenAI’s own API. Free backup: gpt-oss, OpenAI’s open-weight model.
The case against
Claude
Anthropic, via Anthropic API
Anthropic’s Claude, trained differently from GPT, so it tends to catch what the first seat misses. Free backup: Llama (Meta).
How to improve it
Gemini
Google, via Google AI Studio
Google’s Gemini Flash: fast and practical, a good fit for turning an argument into next steps.
Chairman
Qwen
Alibaba, via Groq
Kept off the panel it judges. Reads all three and commits to one verdict.
// Compare two models
Pick any two models to see where each one is stronger. Bars are Council’s editorial ratings out of 10, built from the benchmarks and reviews listed below.
// Benchmarks
The models people actually choose between in October 2026, and where each one sits in Council.
| Model | Lab | In Council | Intelligence Index | Price per 1M tokens (in / out) | Context | Best at |
|---|---|---|---|---|---|---|
| Claude Opus 5.5 | Anthropic | Not in Council yet | 57.6 | $4 / $20 | 1M | Coding and long-form writing; #1 on the Intelligence Index |
| Claude Sonnet 5.5 | Anthropic | Main panel: the case against | 56.0 | $2 / $10 | 1M | Best value at the frontier; writing and following instructions |
| Claude Fable 5.1 | Anthropic | Not in Council yet | 53.4 | $10 / $50 | 1M | The hardest coding and agentic work |
| Claude Haiku 4.5 | Anthropic | Not in Council yet | — | $1 / $5 | 200K | Fast, low-cost everyday tasks |
| GPT-6 Astra | OpenAI | Not in Council yet | 52.7 | $10 / $50 | — | OpenAI’s flagship for hard problems |
| GPT-6.1 Sol | OpenAI | Main panel: the case for | 51.8 | $2 / $10 | — | All-rounder with the largest ecosystem |
| GPT-6 Sol | OpenAI | Main panel (fallback version) | 47.5 | $2 / $10 | 1.05M | Cheap, capable GPT-6 coding |
| GPT-6 Luna | OpenAI | Not in Council yet | 35.0 | $0.10 / $0.50 | — | The lowest-cost OpenAI model for high-volume work |
| Gemini 4 Argon | Not in Council yet | 52.6 | — | — | Reasoning and research at a lower cost per task | |
| Gemini 3.1 Pro | Not in Council yet | — | $2 / $12 | 1M | Long documents and research | |
| Gemini Flash | Main panel: how to improve it | — | Free tier | 1M | Fast bulk tasks | |
| DeepSeek V4 Pro | DeepSeek | Not in Council yet | — | Open weight | — | Low-cost reasoning: 61.3% on ARC-AGI-2 at $0.60 a task |
| DeepSeek V4 Flash | DeepSeek | Not in Council yet | — | Open weight | — | 61.4% on ARC-AGI-2 at $0.04 a task |
| Qwen3.8-Max | Alibaba | Chairman runs an open Qwen model | — | — | — | Cost-effective coding |
| gpt-oss-120b | OpenAI | Free backup | — | Free via Groq | — | Open-weight reasoning, very fast on Groq |
| Llama 4 | Meta | Free backup | — | Open weight | — | Self-hosting and privacy |
| Grok 4 series | xAI | Not in Council yet | — | — | — | Real-time news through X |
Intelligence Index: Artificial Analysis, as reported in October 2026 (higher is better). Prices are list prices per million input and output tokens. A dash means we could not confirm the figure. Highlighted rows are the models Council runs today.
// Planned for Plus
Not live yet: today every question runs on the free panel above. On Plus, Council will sort your question into one of eight types and seat the models suited to it. Type a question to preview the panel it would get.
| Question type | Picked when you ask about | The case for | The case against | How to improve it | Chairman |
|---|---|---|---|---|---|
| Current events | today, this week, latest, news | Grok 4.3 | Gemini 3 Pro | GPT-5.6 Sol | Claude Opus 5.5 |
| Money and investing | investing, loans, rent or buy, tax | GPT-5.6 Sol | DeepSeek V4 Pro | Gemini 3 Pro | Claude Opus 5.5 |
| Career and life decisions | jobs, degrees, moving, relationships | GPT-5.6 Sol | Claude Opus 5.5 | Gemini 3 Pro | GPT-6 Astra |
| Business and startups | startups, launches, pricing, customers | GPT-5.6 Sol | Claude Opus 5.5 | GPT-6 Astra | Gemini 3 Pro |
| Code and technical choices | code, frameworks, databases, hardware | Qwen3.8-Max | Claude Opus 5.5 | GPT-6 Astra | GPT-5.6 Sol |
| Research and long reading | papers, evidence, reports, sources | Gemini 3 Pro | Claude Opus 5.5 | Kimi K2 | GPT-5.6 Sol |
| Writing and creative work | names, essays, emails, slogans | Claude Opus 5.5 | GPT-5.6 Sol | Mistral Large 3 | Gemini 3 Pro |
| General judgment | anything else | GPT-5.6 Sol | Claude Opus 5.5 | Gemini 3 Pro | DeepSeek V4 Pro |
These panels are planned, not live. The free preview always runs the free panel above, and every answer shows exactly which model wrote it.
// Side by side
The five most people actually choose between.
| Model | Best at | Coding | Writing | Reasoning | Live web | Cost |
|---|---|---|---|---|---|---|
| GPT-6 Astra | Agentic, multi-step work | Excellent | Very good | Excellent | Native | Premium |
| Claude Opus 5.5 | Writing, long documents | Excellent | Excellent | Very good | Native | High |
| GPT-5.6 Sol | General reasoning | Excellent | Very good | Excellent | Native | High |
| Gemini 3 Pro | Research, huge context | Very good | Very good | Excellent | Native | Mid |
| Grok 4.3 | Current events | Good | Good | Very good | Native (X) | Low |
// One model or a panel
Council is slower and costs more. If the question is easy, use one model.
| One model | Council | |
|---|---|---|
| Answers | One | Three, then one verdict |
| The case against | Rarely | Always, it is a seat |
| Disagreement | Hidden | Shown under the verdict |
| Confidence | Same tone for everything | High, medium or low |
| Which model | One lab | Three labs on every question |
| Speed and cost | Fast, cheap | Slower, dearer |
Strengths are drawn from public comparison charts, not our own testing. We do not build models. We call them through their providers’ APIs and put them into a structured argument. What we built is the argument.