Model finder
Which AI models should you test? Five questions, up to three models, a monthly cost estimate and a test plan you can run this week.
Model finder
Which AI models should you test?
Five quick questions. We rule out what cannot work, then give you up to three models to test, with an estimated monthly cost, the reasons, what we do not know, and a test plan.
No sign-up. Nothing leaves your browser.
How the finder works
- What are you building?Coding, a support assistant, research, data extraction or agents. This sets the test plan you get at the end.
- What must it handle?Images, audio, video or tool calling. Any model that lacks an input you need is removed.
- How much must it read at once?Your longest typical input, from a few pages to a whole repository. Models with a smaller context window are removed.
- How much will it run?Requests a month, size of a typical request and a budget ceiling. We turn list prices into an estimated monthly bill.
- Any deployment constraint?Open weights only, if you need to self-host or fine-tune. Check each licence before commercial use.
Then it removes every model that fails a requirement, estimates a monthly cost for the rest, and shows up to three in order of cost. Each result lists why it fits and what we do not know. Your answers stay in your browser, and the share link holds them in the address only.
Models the finder screens
| Model | Provider and licence | Inputs | Context | USD per 1M tokens (in / out) |
|---|---|---|---|---|
| Claude Fable 5.1 | Anthropic, closed | text, image | 1M | $10 / $50 |
| Claude Opus 5.5 | Anthropic, closed | text, image | 1M | $4 / $20 |
| Claude Sonnet 5.5 | Anthropic, closed | text, image | 1M | $2 / $10 |
| DeepSeek V4.1 Flash | DeepSeek, open weights | text, image | 1M | $0.3 / $1.2 |
| Gemini 3.8 Flash | Google, closed | text, image, audio, video, tools | Not supplied | $0.75 / $3.75 |
| GLM-5.3 | Zhipu (Z.ai), open weights | text | Not supplied | $1.4 / $4.4 |
| GPT-6 Astra | OpenAI, closed | text, image | 1.05M | $10 / $50 |
| GPT-6 Luna | OpenAI, closed | text, image | 1.05M | $0.1 / $0.5 |
| GPT-6.1 Sol | OpenAI, closed | text, image | 1.05M | $2 / $10 |
| Grok 4.7 | xAI, closed | text, image, tools | 500K | $2 / $6 |
| Kimi K3 | Moonshot AI, open weights | text, image, video | 1M | n/a / n/a |
| Mistral Medium 3.5 | Mistral, closed | text, image, tools | Not supplied | n/a / n/a |
| Muse Spark 1.3 | Meta, closed | text, image | Not supplied | n/a / n/a |
| Qwen3.8 Max | Alibaba/Qwen, closed | text, image, video | Not supplied | n/a / n/a |
“n/a” and “Not supplied” mean no figure was supplied. We never fill a gap with a guess, and a model with no price cannot satisfy a budget limit.
What the finder does not do
Frequently asked questions
How does the Model Finder choose models?
It removes every model that fails a requirement you set: a missing input type, a context window that is too small, a monthly cost over budget, or a licence you cannot use. The remaining models are shown in order of estimated monthly cost.
Is the Model Finder a ranking of the best AI models?
No. No quality benchmark for these models is published on this site, so the finder cannot say which is better. It tells you which models fit your constraints and are worth testing on your own prompts.
Where do the prices and context sizes come from?
They are list prices and context windows supplied by the editorial team on 2026-10-14. We have not yet re-verified each one against the vendor page, and a missing figure is shown as unknown rather than estimated.
How is the monthly cost estimated?
Requests a month multiplied by your typical input and output tokens at each model's list price per million tokens. It leaves out caching, reasoning tokens, batch discounts and taxes, so treat it as a rough guide.
Does the Model Finder store my answers?
No. It runs in your browser and sends nothing to a server. A share link encodes your answers in the web address, so anyone with the link can see them.
Does an affiliate or sponsor influence the shortlist?
No. The finder uses only your answers and the model data. It contains no affiliate links and no sponsored placement.