The Artificial Analysis chart
Gemini 3.1 Flash Lite is the fastest of the models we track at 365 tokens/s of output, measured by Artificial Analysis, followed by Claude Haiku 5.5 (238 tokens/s) and GPT-5.4 Mini (233 tokens/s).
Run current Claude, GPT, Gemini and open models in your agent from $19/month
See plans| Model | Output speedArtificial Analysis | Cost per requestour data |
|---|---|---|
| 1. Gemini 3.1 Flash Lite | 365 tokens/s | $0.0082 |
| 2. Claude Haiku 5.5max effort | 238 tokens/s | $0.013 |
| 3. GPT-5.4 Minixhigh effort | 233 tokens/s | $0.023 |
| 4. Qwen3.7 Max | 197 tokens/s | $0.063 |
| 5. DeepSeek V4.1 Flashmax effort | 195 tokens/s | $0.0067 |
| 6. Gemini 3.8 Flashhigh effort | 145 tokens/s | $0.048 |
| 7. Gemini 3.1 ProGemini 3.1 Pro Preview | 144 tokens/s | $0.095 |
| 8. Claude Sonnet 5.5max effort | 136 tokens/s | $0.053 |
| 9. GPT-6 Lunamax effort | 136 tokens/s | $0.0041 |
| 10. GPT-5.6 Terramax effort | 135 tokens/s | $0.084 |
| 11. MiniMax M3 | 119 tokens/s | $0.013 |
| 12. Muse Wondermax effort | 115 tokens/s | $0.11 |
| 13. GPT-5.3-Codexxhigh effort | 113 tokens/s | $0.055 |
| 14. DeepSeek V4 ProDeepSeek V4 Pro 0813, max effort | 99 tokens/s | $0.030 |
| 15. Grok 4.20Grok 4.20 0309 v2, reasoning | 95 tokens/s | $0.068 |
| 16. Claude Opus 5.5max effort | 95 tokens/s | $0.11 |
| 17. GPT-5.4xhigh effort | 95 tokens/s | $0.10 |
| 18. Step 5 Preview | 90 tokens/s | $0.025 |
| 19. Grok 4.7xhigh effort | 86 tokens/s | $0.13 |
| 20. Kimi K2.7 Code | 75 tokens/s | $0.027 |
| 21. Claude Fable 5.1max effort | 71 tokens/s | $0.24 |
| 22. GLM 5.3max effort | 69 tokens/s | $0.057 |
| 23. MiniMax M2.1 | 67 tokens/s | $0.0065 |
| 24. GPT-6.1 Solmax effort | 65 tokens/s | $0.067 |
| 25. GPT-6 Astramax effort | 63 tokens/s | $0.41 |
| 26. Qwen3.7 Plus | 55 tokens/s | $0.014 |
| 27. Qwen3 235B A22BQwen3 235B A22B 2507 Instruct | 48 tokens/s | $0.0096 |
| 28. Llama 4 Maverick | 44 tokens/s | $0.0095 |
| 29. Kimi K2.6reasoning | 43 tokens/s | $0.024 |
| 30. GLM 4.7reasoning | 42 tokens/s | $0.017 |
| 31. GLM 4.6reasoning | 36 tokens/s | $0.015 |
| 32. DeepSeek R1 (0528)may '25 | 28 tokens/s | $0.031 |
Source: Artificial Analysis (artificialanalysis.ai). Each model is measured as the variant named under it. Cost per request is ours: every one of 79,301 real coding-agent requests priced at list rates (OpenRouter's listed price for some older and open models). Methodology
| Model | Output speedArtificial Analysis | Cost per requestour data |
|---|---|---|
| Gemini 3.5 Flashsuperseded by Gemini 3.8 Flash | 197 tokens/s | $0.049 |
| GPT-5.6 Lunasuperseded by GPT-6 Luna | 123 tokens/s | $0.0084 |
| GPT-6 Solsuperseded by GPT-6.1 Sol | 123 tokens/s | $0.083 |
| Grok 4.3superseded by Grok 4.7 | 113 tokens/s | $0.068 |
| GPT-5.5superseded by GPT-6.1 Sol | 105 tokens/s | $0.21 |
| Claude Haiku 4.5superseded by Claude Haiku 5.5 | 95 tokens/s | $0.022 |
| GPT-5.6 Solsuperseded by GPT-6.1 Sol | 94 tokens/s | $0.17 |
| GLM 5.2superseded by GLM 5.3 | 83 tokens/s | $0.057 |
| Mistral Large 3superseded by Mistral Large 4 | 73 tokens/s | $0.012 |
| Claude Fable 5superseded by Claude Fable 5.1 | 67 tokens/s | $0.32 |
| Claude Opus 4.8superseded by Claude Opus 5.5 | 62 tokens/s | $0.16 |
| Claude Sonnet 4.6superseded by Claude Sonnet 5.5 | 57 tokens/s | $0.097 |
Source: Artificial Analysis (artificialanalysis.ai). Cost per request is ours. Methodology
Output speed is measured on one provider per model. The same open-weight model can be several times faster or slower on another host, and speed drops on very long prompts.
Current Claude, GPT, Gemini and open models are included with Standard Compute
