Price against quality, where new models come from, open weights against proprietary, and who serves the most tokens on OpenRouter. Built from our own data and recomputed every 10 minutes.
Sources: OpenRouter's public catalog and rankings, our human-vote battles and chess arena. How these charts are made → /api/v1/charts
Fig 01 · Price vs quality
Price vs quality
x: blended price per 1M tokens at 3 input : 1 output, log scale. y: rating. Free models are left out.
Only 0 judged battles so far — too few to place models by human votes. Each vote you cast adds to this chart.
Vote in a battle →Google Gemini 3.7 Flash rates highest; Anthropic Claude Haiku 5.5 is the cheapest model on the frontier at $0.20 per 1M tokens, 7.5× cheaper.
On the frontier, cheapest first
| Model | Rating | Range | Votes / games | Blended price | On frontier |
|---|---|---|---|---|---|
| Google Gemini 3.7 Flash | 2162 | 1929–2395 | 31 games | $1.50 | yes |
| SpaceXAI Grok 4.3 | 1693 | 1522–1864 | 31 games | $1.56 | |
| Anthropic Claude Haiku 5.5 | 1655 | 1486–1824 | 31 games | $0.20 | yes |
| Z.ai GLM 5.3 Flash | 1497 | 1326–1668 | 31 games | $0.237 | |
| Qwen3.8 Flash | 1473 | 1305–1641 | 31 games | $0.23 | |
| DeepSeek V4.1 Flash | 1468 | 1310–1626 | 33 games | $0.21 | |
| Mistral Small 4 | 1341 | 834–1849 | 1 games | $0.262 | |
| MoonshotAI Kimi K2.6 | 1228 | 1061–1395 | 30 games | $1.34 | |
| MiniMax M3.1 Flash | 1024 | 844–1203 | 31 games | $0.262 |
Fig 02 · Countries
Each lab is mapped to the country of its headquarters. No Russian lab (Yandex, Sber GigaChat, T-Bank) is on OpenRouter yet, so Russia doesn't appear here.
New models by country
Models added to OpenRouter per quarter, by the lab's home country
US: 50% of new models in Q4 2026 so far, more than any other country.
| Quarter | Total | US | China | France | Canada | South Korea | Japan | Other |
|---|---|---|---|---|---|---|---|---|
| Q1 2025 | 21 | 12 | 3 | 3 | 1 | 0 | 0 | 2 |
| Q2 2025 | 19 | 13 | 4 | 2 | 0 | 0 | 0 | 0 |
| Q3 2025 | 30 | 13 | 15 | 1 | 0 | 0 | 0 | 1 |
| Q4 2025 | 33 | 22 | 11 | 0 | 0 | 0 | 0 | 0 |
| Q1 2026 | 43 | 22 | 18 | 1 | 0 | 1 | 0 | 1 |
| Q2 2026 | 42 | 22 | 17 | 1 | 1 | 0 | 1 | 0 |
| Q3 2026 | 85 | 44 | 26 | 0 | 1 | 2 | 3 | 9 |
| Q4 2026 | 26 | 13 | 4 | 1 | 0 | 0 | 0 | 8 |
Best-rated model per country
Chess arena · current ratings; rating history isn't kept yet
The top-rated model is from US: Google Gemini 3.7 Flash. Best from China: Z.ai GLM 5.3 Flash.
Labs by country
Current catalog: models, share with open weights, median blended price per 1M tokens, best-rated model
26% of US's models have open weights at a median $1.50; China: 75% at $0.471.
US174 models
Open weights 26% · Median price $1.50
Google Gemini 3.7 Flash2162China114 models
Open weights 75% · Median price $0.471
Z.ai GLM 5.3 Flash1497France20 models
Open weights 55% · Median price $0.426
Mistral Small 41341Canada6 models
Open weights 33% · Median price $0.60
Japan4 models
Open weights 0% · Median price $7.13
South Korea3 models
Open weights 0% · Median price $0.262
Other28 models
Open weights 50% · Median price $0.731
| Country | Models | Open weights | Median price | Best rated |
|---|---|---|---|---|
| US | 174 | 26% | $1.50 | Google Gemini 3.7 Flash2162 |
| China | 114 | 75% | $0.471 | Z.ai GLM 5.3 Flash1497 |
| France | 20 | 55% | $0.426 | Mistral Small 41341 |
| Canada | 6 | 33% | $0.60 | — |
| Japan | 4 | 0% | $7.13 | — |
| South Korea | 3 | 0% | $0.262 | — |
| Other | 28 | 50% | $0.731 | — |
Fig 03 · Open weights
A model counts as open-weights when OpenRouter links its weights on Hugging Face; everything else is proprietary.
Models added to OpenRouter per quarter
| Quarter | Open weights | Proprietary | Not classified | Share |
|---|---|---|---|---|
| Q1 2025 | 11 | 10 | 0 | 52% |
| Q2 2025 | 7 | 12 | 0 | 37% |
| Q3 2025 | 19 | 11 | 0 | 63% |
| Q4 2025 | 11 | 21 | 1 | 34% |
| Q1 2026 | 15 | 28 | 0 | 35% |
| Q2 2026 | 17 | 24 | 1 | 41% |
| Q3 2026 | 31 | 54 | 0 | 36% |
| Q4 2026 | 11 | 15 | 0 | 42% |
Best open vs best proprietary
Chess arena · current ratings
The best open-weights model, Z.ai GLM 5.3 Flash, is 665 points behind the best proprietary one, Google Gemini 3.7 Flash.
Median price
Blended $ per 1M tokens, paid text models in the catalog
A typical proprietary model costs 5.8× as much as a typical open-weights one: $1.98 against $0.342 per 1M tokens.