AI models, compared on what we've checked.
Each profile separates three things: what the provider says, what we've verified against their documentation, and what we've found by testing. You'll always see which is which.
Model families
| Family | Provider | Current models | Context window | Open weights | Sources checked |
|---|---|---|---|---|---|
| Claude | ProviderAnthropic | Current models4 | Context windowVerified. 1M tokens | Open weightsDraft. No | Sources checked 11 Nov 2026 |
| GPT | ProviderOpenAI | Current models4 | Context windowVerified. 1.05M tokens | Open weightsDraft. No | Sources checked 11 Nov 2026 |
| Gemini | ProviderGoogle | Current models6 | Context windowVerified. 1M tokens | Open weightsDraft. No | Sources checked 11 Nov 2026 |
| Grok | ProviderxAI | Current models1 | Context windowVerified. 500K tokens | Open weightsDraft. No | Sources checked 11 Nov 2026 |
| Llama | ProviderMeta | Current models3 | Context windowNot yet checked. Not yet checked | Open weightsDraft. Yes | Sources checked 11 Nov 2026 |
Checked against the provider's pageOur draft, not yet checkedNot yet checked
Put two side by side
Only facts we've checked are shown as facts. Gaps stay visible instead of being filled with guesses.
- Provider
- Verified. Anthropic
- Verified. OpenAI
- Current models
- Verified. Claude Fable 5.2, Claude Opus 5.5, Claude Sonnet 5.5, Claude Haiku 5.5
- Verified. GPT-6 Astra, GPT-6.1 Sol, GPT-6.1 Terra, GPT-6 Luna
- Context window
- Verified. 1M tokens (all four current models)
- Verified. 1.05M tokens (all four GPT-6 models)
- Max output
- Verified. 128K tokens
- Verified. 128K tokens
- Open weights
- Draft. No
- Draft. Not for the models profiled here
- API pricing
- Verified. From $0.10 in / $0.50 out (Haiku 5.5) to $8 in / $40 out (Fable 5.2), per million tokens
- Verified. From $0.10 in / $0.50 out (GPT-6 Luna) to $10 in / $50 out (GPT-6 Astra), per million tokens
- Developer access
- Not yet checked. Claude API. Cloud platforms such as Amazon Bedrock and Google Cloud: not re-checked: not yet checked
- Verified. OpenAI API
- Full profile
- Full Claude profilePer-model prices and detailsVerified facts from Anthropic: Models overview (opens in a new tab), checked 12 Nov 2026Verified facts from Claude: Plans and pricing (opens in a new tab), checked 12 Nov 2026
- Full GPT profilePer-model prices and detailsVerified facts from OpenAI: Models (opens in a new tab), checked 12 Nov 2026Verified facts from OpenAI: GPT-6 Astra (opens in a new tab), checked 12 Nov 2026Verified facts from OpenAI: API pricing (opens in a new tab), checked 12 Nov 2026
How a profile is built
Step 1
Read the source
We take lineups, limits and prices from the provider's own documentation, and record the page and the date.
Step 2
Run the same tests
Every model gets the same four tasks for coding, writing, research and reasoning, so results are comparable.
Step 3
Re-check on a schedule
Model pages change often. Each profile shows when its sources were last checked so you can judge how fresh it is.
Our first head-to-head run is in model matchups. Profiles don’t include our own findings yet. Read our standards