About Toolprism
Toolprism tracks what every large language model costs, which provider serves it cheapest, and how the models actually score against each other — alongside hand-written reviews of the AI tools built on top of them.
Why it exists
Model pricing is scattered across dozens of vendor pages, quoted in different units, and changes without notice. The same open-weight model can cost three times more depending on who is serving it. Benchmark numbers live in leaderboards that are hard to line up against price. This site puts those three things — price, capability and provider — on one page for each of the 340 models currently callable, and keeps them current automatically.
Where the data comes from
- OpenRouter model API — model catalogue, per-provider pricing, context windows, capability flags, uptime and throughput.
- Artificial Analysis — intelligence, coding and agentic index scores (via the OpenRouter model API).
- Design Arena — head-to-head win rates and Elo by build category (via the OpenRouter model API).
- Product Hunt — the AI tool catalogue and launch metadata.
- Vendor pricing pages — plan pricing for the tools we review by hand.
The model catalogue is pulled from the OpenRouter public API and written into the site's database, then the pages are regenerated as static HTML. Nothing on a model page is typed by hand, which is the point: when a model launches or a provider cuts its price, the change appears on the next rebuild rather than whenever an editor gets round to it. Every page shows the date it was last rebuilt. Last rebuild: August 3, 2026.
How the rankings are computed
Each of the 15 task rankings states the rule it sorted by, at the top of the page, before the list. A ranking either filters on a capability the API reports (tool calling, structured output, image input, published weights, context window) or sorts on a published score (Artificial Analysis intelligence, coding and agentic indices; Design Arena win rates). Where a model has no score, it does not appear in a ranking that depends on one — an unscored model is left out rather than guessed at.
We do not run our own evaluations, and we do not publish a subjective "best of" list. If you disagree with a ranking, the number it sorted by is on the page and you can check it against the source.
How this site makes money
Some links to tools and services are affiliate links: if you sign up through one, we earn a commission at no extra cost to you. Those relationships are with Impact, PartnerStack, direct vendor programs. They do not affect rankings — the model rankings are computed from public data and there is no affiliate program in the model layer at all. See the affiliate disclosure for the full picture.
What is here
- 340 model pages — pricing per 1M tokens, every provider, benchmarks, capabilities.
- LLM API pricing table — the whole catalogue sorted by cost, with a workload calculator.
- 15 task rankings — coding, agents, reasoning, long context, vision, cheapest, free, self-hosting and more.
- Model head-to-heads — two models side by side on price, context and scores.
- 1003 AI tools — catalogue with hand-written reviews on the tools we know well.
Corrections and contact
If a price is wrong, a model is missing, or a ranking looks off, tell us and we will fix it — write to [email protected]. Vendors who want a listing corrected can use the same address; we do not sell placement in the rankings, but we do fix factual errors quickly.