AI models, in context.
If you are choosing a model for an application, start from the job and the budget, then read the profiles in one family. Each vendor sells a range: a most capable model, a balanced middle tier and a fast, low-cost tier.
| Compare | Model | Provider | Modalities | Context | Input price | HumanRank result |
|---|---|---|---|---|---|---|
| Claude Fable 5.1Source-reviewed profile | Anthropic | Text and images in; text out (stated for all current models) | 1M tokens | Check profile | Not tested | |
| Claude Haiku 5.5Source-reviewed profile | Anthropic | Text and images in; text out | 1M tokens | Check profile | Not tested | |
| Claude Opus 5.5Source-reviewed profile | Anthropic | Text and images in; text out | 1M tokens | Check profile | Not tested | |
| Claude Sonnet 5.5Source-reviewed profile | Anthropic | Text and images in; text out | 1M tokens | Check profile | Not tested | |
| DeepSeek V4.1 FlashSource-reviewed profile | DeepSeek | Check profile | Check profile | Check profile | Not tested | |
| DeepSeek V4 ProSource-reviewed profile | DeepSeek | Check profile | Check profile | Check profile | Not tested | |
| Gemini 3.5 Flash-LiteSource-reviewed profile | Check profile | Check profile | Check profile | Not tested | ||
| Gemini 3.8 FlashSource-reviewed profile | Check profile | Check profile | Check profile | Not tested | ||
| GPT-6.1 SolSource-reviewed profile | OpenAI | Text and image in; text out; audio and video unsupported | 1,050,000 tokens (maximum input 922,000 tokens) | USD 2.00 per 1M tokens (list price as published on 15 October 2026) | Not tested | |
| GPT-6 AstraSource-reviewed profile | OpenAI | Text and image in; text out | 1,050,000 tokens (maximum input 922,000 tokens) | USD 10.00 per 1M tokens (list price as published on 15 October 2026) | Not tested | |
| GPT-6 LunaSource-reviewed profile | OpenAI | Text and image in; text out | 1,050,000 tokens (maximum input 922,000 tokens) | USD 0.10 per 1M tokens (list price as published on 15 October 2026) | Not tested | |
| Grok 4.7Source-reviewed profile | Grok | Text and image in; text out | 500,000 tokens | USD 2.00 per 1M tokens below 200k prompt tokens; USD 4.00 at or above (list price as published on 15 October 2026) | Not tested |
List prices and limits are date-stamped in each profile. Verify with the provider before budgeting.
Compare model profiles
Choose two or three models to compare.
What each profile covers
Every profile uses the same sections, so you can compare models side by side:
- Key specifications: the exact API identifier, input and output types, context window, maximum output, knowledge cutoff and reasoning controls.
- List price: the vendor's published price per million tokens in US dollars, with any long-prompt surcharges, caching rates, batch discounts or time-of-day pricing the vendor documents.
- When to use it: what the vendor says the model is for, followed by our editorial reading, labelled as such.
- Limits and costs to plan for: unsupported features, rate limits, regional processing and scheduled price changes.
- How to evaluate it: a short test you can run on your own work before committing.
How to choose between families
Start with your constraints. If your data must stay in a region, check data residency first, because several models offer it only on certain endpoints. If latency matters, compare the vendor's own latency labels and test with your prompts. If cost dominates, compare the price of the full request, including long prompts and cached input, not just the headline rate.
Pin the exact version. Vendors use aliases that can move to newer versions. DeepSeek's documentation, for example, says retired model names are now served by a newer model. Record the identifier and date in every evaluation so a result can be repeated.
Test the middle tier first. A balanced model is often good enough, and its result tells you whether the most capable tier is worth several times the price.
Watch dated price changes. Google's pricing page announces that Gemini 3.8 Flash token prices change on 1 January 2027. Check the vendor page before you budget.
Where models meet products
Most people use these models through products rather than APIs. Our guides to AI chatbots and AI coding tools cover the apps built on top of them, including which models each product lets you select.
Method and limits
We read each vendor's model page, pricing page and selection guide on 15 October 2026: the OpenAI model list, the Claude models overview, the Gemini API model list, DeepSeek's models and pricing page and the Grok models page. Every figure in a profile links to its source. Prices and limits change often, and some vendors run promotions or scheduled changes. We have not run controlled tests, so no model has a HumanRank score yet. Our ranking method explains what a test must include before we publish one.