Every AI Provider · Tracked
The most complete list of AI providers you can actually use · 273 providers across 8 countries, shipping 1,057 models · ranked by model count, pricing, and average benchmark score.
The Value Quadrant
Quality vs price · one chart, every provider
Top 10 Providers
Ranked by model count weighted by average benchmark score
What's moving
Volume leaders, quality champions, OSS heroes, price leaders
All providers
Filter, sort, and compare · every provider in the database
All Providers
| # | Provider | Country | Models | Open | Open % | Avg Score | From $/M | Top Model | |
|---|---|---|---|---|---|---|---|---|---|
| 1 | 🇫🇷France | 32 | 32 | 100% | 29.3% | $0.02 | Mixtral 8x7B |
1 rows · click column headers to sort
Top 5 by region
Regional leaders across every market we track
Frequently asked
Quick answers, sourced from our data
How many AI providers does BenchGecko track?
BenchGecko currently tracks 273 AI providers across 8 countries, shipping a combined 1,057 models. New providers are added as they launch and the full dataset refreshes daily.
Which AI provider has the most models?
Use the "Most Models" tab in the Movers section above to see current volume leaders. The default BenchGecko rank, however, is quality-weighted: rank = avgScore × log10(modelCount + 1) × 10. That formula rewards providers that ship many high-scoring models and prevents a provider with 200 forks of one base model from outranking one with a smaller, sharper portfolio.
How is the "avg score" calculated per provider?
Provider average score is the arithmetic mean of every benchmark score across every model the provider ships, using BenchGecko's cross-benchmark normalization. Providers with fewer than 2 scored models are excluded from the "Highest Rated" list and the default podium rank, to avoid single-model outliers.
Why are Qwen and Alibaba merged into one provider?
Upstream data sources (OpenRouter, Hugging Face) sometimes emit multiple slugs for the same underlying company · e.g. "qwen" and "alibaba", or "meta", "meta-llama", and "facebook". BenchGecko canonicalizes these into a single entry so model counts and rankings reflect the real corporate structure. See the methodology page for the full alias list.
What counts as an "open source" provider?
Any provider with at least one model that ships weights under a permissive or source-available license. We count Llama, Qwen, DeepSeek, Mistral Open, and similar weights-available families as OSS even when the license has usage restrictions. Toggle the "Has Open Source" filter above to see only these providers.
Where do the provider stats come from?
Provider metadata and model inventories come from OpenRouter's public API. Benchmarks are pulled from Epoch AI (CC-BY) and SWE-bench's public leaderboards. Country and founding-year data is compiled by BenchGecko. See the methodology page for full provenance.
Can I filter by region?
Yes. The region pills at the top of the filter bar let you narrow to US & Canada, Europe, China, or the rest of the world. Region is based on HQ country, which is sourced from official filings and provider websites.
Why do some providers show "—" for cheapest price?
Some providers only list their models as "Free" (via limited preview endpoints) or have not yet published public per-token pricing. These show as "—" until we have a non-zero price to display.
How do I navigate to a provider page?
Click any provider name anywhere on this page — the podium, the movers table, the all-providers table, or the regional mini tables. Each provider has a dedicated page with its full model list, pricing, OSS ratio, and category breakdown.