Model rankings
The coding models developers actually use on Tokens, ranked by tokens processed. Real traffic, updated through the day.
The last 7 days
- Tokens processed
- 13.4B+140.6% vs the 7 days before
- Models in use
- 46each used by 1+ developers
Daily usage
Tokens per day over the last 30 days, split across the most used models.
Tokens per day, by model
21.6B tokens in 30 days
Hover, touch or use the arrow keys to read a day.
Leaderboard: this week
Ranked by tokens. Arrows show how a model moved since the period before.
- Requests
- 57.1K
- Change
- +87.7%
- Requests
- 29.4K
- Change
- +285.2%
- Requests
- 11.7K
- Change
- +628.3%
- Requests
- 12.5K
- Change
- +497.7%
- Requests
- 18.6K
- Change
- −29.9%
- 62 places down
DeepSeek V4.1 Flash Fastdeepseek/deepseek-v4.1-flash-fast48.7M
0.4%
- Requests
- 462
- Change
- −65.5%
- Requests
- 419
- Change
- +12,881%
- Requests
- 766
- Change
- +194.6%
- Requests
- 228
- Change
- +201,770%
- Requests
- 1.45K
- Change
- +1,488%
- Requests
- 626
- Change
- +13,087%
- Requests
- 46
- Change
- +3,657,060%
- 135 places down
DeepSeek V4 Flash Fastdeepseek/deepseek-v4-flash-fast22.8M
0.2%
- Requests
- 196
- Change
- −29.5%
- Requests
- 376
- Change
- +193.5%
- Requests
- 205
- Change
- +2,471%
- Requests
- 221
- Change
- +283,021%
- Requests
- 174
- Change
- +1,541,870%
- Requests
- 139
- Change
- +15.1%
- Requests
- 44
- Change
- +1,074,340%
- 2016 places up
Grok 4.5x-ai/grok-4.52.55M
0%
- Requests
- 940
- Change
- +38,725%
- Requests
- 46
- Change
- +42.7%
- Requests
- 35
- Change
- −6.5%
- Requests
- 18
- Change
- −93%
- Requests
- 16
- Change
- +16,961%
- Requests
- 189
- Change
- −76%
Trending
The fastest-growing models since the period before.
GLM-5.3 Flash1.07B tokens+628.3%
Claude Haiku 5.5578M tokens+497.7%
DeepSeek V4.1 Flash4.23B tokens+285.2%
DeepSeek V4.1 Flash 10177.07B tokens+87.7%
Share by vendor
Which makers' models handle the most tokens.
- Deepseek-ai52.8% · 7.07B
- DeepSeek32.1% · 4.3B
- Z.ai8.2% · 1.1B
- Claude4.6% · 614M
- OpenAI0.7% · 91.3M
- Qwen0.6% · 85.2M
- MiniMax0.4% · 48.2M
- Xiaomi0.3% · 44.1M
- Meta0.2% · 22.9M
- Moonshot AI0.1% · 7.41M
- xAI0% · 3.41M
- inclusionAI0% · 674K
- Google0% · 536K
- Meituan0% · 514K
- StepFun0% · 429
What the tokens are
Reading, writing and cached tokens for the most used models.
- Input
- Output
- Cache
- DeepSeek V4.1 Flash 10175.8% in · 0.7% out · 93.5% cache
- DeepSeek V4.1 Flash1.8% in · 0.4% out · 97.8% cache
- GLM-5.3 Flash34.7% in · 0.9% out · 64.4% cache
- Claude Haiku 5.562.7% in · 1.3% out · 36% cache
GPT-6 Luna5.8% in · 3.2% out · 91% cache- DeepSeek V4.1 Flash Fast1.1% in · 0.5% out · 98.4% cache
MiniMax M34.4% in · 0.6% out · 95% cache
Claude Sonnet 5.511.6% in · 3.8% out · 84.6% cache
How these numbers are counted
- Tokens are input, output and cache tokens added together, for every request made through the API or the playground. Periods are whole UTC days ending today.
- Change compares the last 7 or 30 days with the same length of time just before. Today and all time have no earlier period to compare with.
- Updates. The statistics are rebuilt about every hour, and this page refreshes a few minutes after that.
Privacy
- Only totals are published. No account, key, prompt, price or payment detail is part of this page.
- A model is listed only when at least 1 different developers used it in the period. Smaller models are counted under “other models”.
- Figures are rounded to three significant figures, so no one customer can be picked out.
Use any of these models with one API key
Create an account, add credit, and point your coding agent at one endpoint.
