LLM rankings
The most-used AI models this week
Token usage across models on OpenRouter, the largest public LLM marketplace.
Source: OpenRouter (openrouter.ai/rankings), as of 2026-10-28.
3 of the top 20 run here for 30% less
Same models through one OpenAI-compatible endpoint, billed below each provider’s list price.
Top 20 this week
| # | Model | Tokens | Change | Price on cheapaimodel | Details |
|---|---|---|---|---|---|
| 1 | DeepSeek V4.1 Flash 1017by DeepSeekOn cheapaimodel37.3T+224% | 37.3T | +224% | 30% off$0.20 / $0.83 | |
| 2 | DeepSeek V4.1 Flashby DeepSeekOn cheapaimodel17T−38% | 17T | −38% | 30% off$0.20 / $0.83 | |
| 3 | GLM 5.3 Flashby Z.aiOn cheapaimodel14.1T+22% | 14.1T | +22% | 30% off$0.10 / $0.34 | |
| 4 | GPT-6 Lunaby OpenAI9.1T+46% | 9.1T | +46% | ||
| 5 | Hy4 previewby Tencent9T+10% | 9T | +10% | ||
| 6 | MiniMax M3.1 Flash (free)by MiniMax8.7T+75% | 8.7T | +75% | ||
| 7 | Claude Opus 5.5by Anthropic8.1T+19% | 8.1T | +19% | ||
| 8 | GLM 5.3by Z.aiOn cheapaimodel5.1T−3% | 5.1T | −3% | 30% off$0.97 / $3.07 | |
| 9 | Nemotron 3 Ultra (free)by NVIDIA5T+11% | 5T | +11% | ||
| 10 | MiMo-V2.6-Flashby Xiaomi4.7T−48% | 4.7T | −48% | ||
| 11 | Step 5 Previewby StepFun4.7T−20% | 4.7T | −20% | ||
| 12 | DeepSeek V4 Flash 0731by DeepSeek4.1T−14% | 4.1T | −14% | ||
| 13 | jev-1.13by typesafe3.1T+5% | 3.1T | +5% | ||
| 14 | Hy3by Tencent3T−4% | 3T | −4% | ||
| 15 | GPT-6.1 Solby OpenAI3T+56% | 3T | +56% | ||
| 16 | Claude Haiku 5.5by Anthropic2.9T+47% | 2.9T | +47% | ||
| 17 | DeepSeek V4 Flash 0423by DeepSeek2.9T+3% | 2.9T | +3% | ||
| 18 | Muse Spark 1.3 Contributorby Meta2.6T−20% | 2.6T | −20% | ||
| 19 | Claude Sonnet 5.5by Anthropic2.5T0% | 2.5T | 0% | ||
| 20 | MiniMax M3.1 Flashby MiniMax2.3TNew | 2.3T | New | ||
Source: OpenRouter (openrouter.ai/rankings), as of 2026-10-28. · CC BY 4.0
Prices per 1M input / output tokens
Every highlighted model works with one API key.
Get an API keyWhat would you pay here?
Pick a model and your monthly volume. We compare the provider’s list price with ours, using the two prices on each model page.
- List price
- $302.40
- cheapaimodel
- $211.03
504M input + 126M output at $0.20 / $0.83 per 1M. Cache reads would lower both.
Top apps on OpenRouter
Apps and agents ranked by tokens this week. The ones with a guide work with cheapaimodel.
- 1Hermes Agent12.8T tokens
- 2Claude Code11.5T tokensNeeds the Anthropic API, not available yet
- 3pi6T tokens
- 4Kilo Code5.4T tokens
- 5DeepSeek Harness2.5T tokens
- 6omp2.4T tokens
- 7Codex2.2T tokensSet up Codex with cheapaimodel →
- 8Freebuff1.8T tokens
- 9OpenClaw1.8T tokensSet up OpenClaw with cheapaimodel →
- 10Strix1.7T tokens
Token share by model maker
Share of all tokens on OpenRouter this week (2026-10-21 – 2026-10-27). Makers outside the top 8 are grouped as Others.
- DeepSeekOn cheapaimodel33.3%
- Z.aiOn cheapaimodel10.7%
- OpenAI8.3%
- Anthropic7.6%
- MiniMax6.5%
- Tencent6.4%
- NVIDIA4.0%
- Xiaomi3.3%
- Others19.9%
Weekly token usage
Last 12 weeks: this week’s top 5 models and everything else.
- DeepSeek V4.1 Flash 1017here
- DeepSeek V4.1 Flashhere
- GLM 5.3 Flashhere
- GPT-6 Luna
- Hy4 preview
- Others
View as table
| Week of | DeepSeek V4.1 Flash 1017 | DeepSeek V4.1 Flash | GLM 5.3 Flash | GPT-6 Luna | Hy4 preview | Others |
|---|---|---|---|---|---|---|
| 2026-08-05 | — | — | — | — | — | 71.5T |
| 2026-08-12 | — | — | — | — | — | 79.8T |
| 2026-08-19 | — | — | — | — | — | 104T |
| 2026-08-26 | — | — | 10T | — | 5.7T | 90.7T |
| 2026-09-02 | — | — | 12.2T | — | 18.7T | 92.1T |
| 2026-09-09 | — | 8T | 11.4T | — | 13.6T | 91.5T |
| 2026-09-16 | — | 17.8T | 18.4T | 72.7B | 12.9T | 85T |
| 2026-09-23 | — | 22T | 11.6T | 4.4T | 8.1T | 107T |
| 2026-09-30 | — | 29.6T | 10.2T | 6.4T | 6.4T | 112T |
| 2026-10-07 | — | 40.5T | 12.9T | 7.1T | 8.1T | 108T |
| 2026-10-14 | 11.5T | 27.3T | 11.6T | 6.2T | 8.2T | 99T |
| 2026-10-21 | 37.3T | 17T | 14.1T | 9.1T | 9T | 102T |
Run 3 top-20 models for 30% less.
One OpenAI-compatible endpoint. Pay as you go from $5, no subscription.
How these rankings work
- What is counted
- Input and output tokens per model, as OpenRouter reports them. Each provider counts with its own tokenizer, so token counts aren’t exactly comparable across providers.
- When it updates
- OpenRouter publishes totals for each UTC day. We refresh every hour, so a finished day shows up here the next morning (UTC).
- Which rows link here
- A row links to cheapaimodel when we serve that model. The price shown is ours, below the model’s list price.
- License
- Rankings data © OpenRouter, published under CC BY 4.0.