Median output speed
independentrun by Artificial Analysisunit: tokens_per_sDisplay with attribution; no redistribution47 modelssource ↗
| # | Model | Score | Observed | |
|---|---|---|---|---|
| 1 | Mercury 2Inception | 702 tok/s | Sep 21, 2026 | |
| 2 | Gemini 3.5 Flash LiteGoogle | 402 tok/s | Sep 21, 2026 | |
| 3 | Gemini 2.5 Flash-LiteGoogle | 328 tok/s | Sep 21, 2026 | |
| 4 | Gemini 3.5 FlashGoogle | 244 tok/s | Sep 21, 2026 | |
| 5 | Muse Spark 1.2Meta | 234 tok/s | Sep 20, 2026 | |
| 6 | Gemini 2.5 FlashthinkingGoogle | 230 tok/s | Sep 21, 2026 | |
| 7 | MiniMax-M3MiniMax (minimax.io) | 227 tok/s | Sep 21, 2026 | |
| 8 | o3-minihigh-effortOpenAI | 224 tok/s | Sep 21, 2026 | |
| 9 | GPT-5.1 CodexOpenAI | 196 tok/s | Jul 25, 2026 | |
| 10 | GPT-4.1OpenAI | 190 tok/s | Sep 21, 2026 | |
| 11 | Qwen3-Next 80B-A3B InstructAlibaba | 178 tok/s | Sep 21, 2026 | |
| 12 | GPT-5.6 LunaOpenAI | 166 tok/s | Sep 21, 2026 | |
| 13 | nemotron-3-nano-30b-a3bNvidia | 160 tok/s | Sep 21, 2026 | |
| 14 | nvidia-nemotron-nano-9b-v2Nvidia | 154 tok/s | Sep 21, 2026 | |
| 15 | GPT-5.2 CodexOpenAI | 152 tok/s | Jul 25, 2026 | |
| 16 | Qwen3.5 35B-A3BAlibaba | 147 tok/s | Sep 21, 2026 | |
| 17 | Gemini 2.5 ProGoogle | 145 tok/s | Sep 21, 2026 | |
| 18 | Qwen3.5 122B-A10Bnon-reasoningAlibaba | 145 tok/s | Sep 21, 2026 | |
| 19 | o3OpenAI | 144 tok/s | Sep 21, 2026 | |
| 20 | GPT-5.5OpenAI | 115 tok/s | Sep 21, 2026 | |
| 21 | GPT-5OpenAI | 113 tok/s | Sep 21, 2026 | |
| 22 | GPT-5.4non-reasoningOpenAI | 111 tok/s | Sep 21, 2026 | |
| 23 | MiniMax-M2.1MiniMax (minimax.io) | 111 tok/s | Sep 21, 2026 | |
| 24 | GPT-5.2OpenAI | 101 tok/s | Sep 21, 2026 | |
| 25 | GPT-5 MiniOpenAI | 100 tok/s | Sep 21, 2026 | |
| 26 | Qwen TurboAlibaba | 100 tok/s | Sep 21, 2026 | |
| 27 | MiniMax-M2.5MiniMax (minimax.io) | 100 tok/s | Sep 21, 2026 | |
| 28 | GPT-5.1OpenAI | 94 tok/s | Sep 1, 2026 | |
| 29 | GLM-5.3-FlashZhipu AI | 94 tok/s | Sep 21, 2026 | |
| 30 | Qwen3.5 397B-A17Bnon-reasoningAlibaba | 91 tok/s | Sep 21, 2026 | |
| 31 | Qwen3.5 27Bnon-reasoningAlibaba | 83 tok/s | Aug 27, 2026 | |
| 32 | GLM-4.7-FlashZhipu AI | 80 tok/s | Sep 21, 2026 | |
| 33 | Claude Sonnet 5Anthropic | 79 tok/s | Sep 21, 2026 | |
| 34 | GLM-5Zhipu AI | 77 tok/s | Sep 21, 2026 | |
| 35 | GLM-4.6VZhipu AI | 75 tok/s | Sep 21, 2026 | |
| 36 | GPT-6 AstraOpenAI | 71 tok/s | Sep 21, 2026 | |
| 37 | Claude Fable 5.1Anthropic | 70 tok/s | Sep 21, 2026 | |
| 38 | Qwen3.6 27Bnon-reasoningAlibaba | 64 tok/s | Sep 21, 2026 | |
| 39 | MiniMax-M2.7MiniMax (minimax.io) | 64 tok/s | Sep 2, 2026 | |
| 40 | Claude Opus 5xhigh-effortAnthropic | 58 tok/s | Sep 21, 2026 | |
| 41 | Grok Build 0.1xAI | 52 tok/s | Sep 21, 2026 | |
| 42 | Claude Opus 4.5 (latest)thinkingAnthropic | 51 tok/s | Sep 21, 2026 | |
| 43 | Qwen3-Coder 480B-A35B InstructAlibaba | 48 tok/s | Sep 21, 2026 | |
| 44 | Claude Opus 4.7non-reasoningAnthropic | 45 tok/s | Sep 21, 2026 | |
| 45 | Claude Sonnet 4.6Anthropic | 44 tok/s | Sep 21, 2026 | |
| 46 | Claude Opus 4.6Anthropic | 42 tok/s | Sep 21, 2026 | |
| 47 | GLM-4.5VthinkingZhipu AI | 41 tok/s | Sep 20, 2026 |
Bars on a 0–702.37 tokens_per_s scale. Scores link to model pages; every number keeps its source on the model page.