Ranking
Median output tokens per second after the first token, over repeated streamed runs of the same prompt. Models that deliver the whole answer in one burst after hidden thinking are left out. Top 15 of 72.
| # | Model | Speed (tok/s) | First token | Score | Cost / question |
|---|---|---|---|---|---|
| 1 |  Nemotron 3.5 Super VLNVIDIA | 715 | 0.6 s | 40.0% | $0.0013 |
| 2 |  Gemini 3.5 Flash LiteGoogle | 409 | 1.1 s | 75.0% | $0.0034 |
| 3 |  Trinity Large ThinkingArcee AI | 329 | 0.6 s | 28.8% | $0.0025 |
| 4 |  Ministral 3 3B 2512Mistral | 252 | 0.3 s | 26.2% | $0.00019 |
| 5 |  Qwen3.5 397B A17BQwen | 240 | 0.6 s | 32.5% | $0.013 |
| 6 |  Inkling SmallThinking Machines | 233 | 0.5 s | 67.5% | $0.00047 |
| 7 |  Muse Spark OpenMeta | 220 | 0.5 s | 77.5% | $0.0045 |
| 8 |  Claude Haiku 5.5Anthropic | 192 | 0.7 s | 68.8% | $0.00048 |
| 9 |  Nova Micro 1.0Amazon | 180 | 0.5 s | 22.5% | $0.000087 |
| 10 | Muse WonderMeta | 175 | 16.1 s | 78.8% | $0.021 |
| 11 | Gemma 4 26B A4BGoogle | 172 | 0.7 s | 15.0% | $0.0011 |
| 12 | Mistral Small 4Mistral | 170 | 0.5 s | 32.5% | $0.0019 |
| 13 |  DeepSeek V4 Flash Vision ExpDeepSeek | 170 | 0.7 s | 71.2% | $0.0011 |
| 14 |  o3OpenAI | 163 | 1.2 s | 72.5% | $0.0088 |
| 15 |  Mercury 2.5Inception | 157 | 2.3 s | 58.2% | $0.00011 |