Leaderboard
Multi-dimensional rankings based on model speed tests and provider health checks. Compare providers, endpoints, and reliability at a glance.
Average time to first token. Lower is better for responsiveness.
| Rank | Provider | Model | First Token Latency | Avg tokens per second | Total Tests |
|---|---|---|---|---|---|
| 1 | qwen/qwen3-coder | 0.05 s Best: -Worst: 1.20 | 0.28t/s | 25 | |
| 2 | deepseek-v3.2 | 0.27 s Best: -Worst: 0.73 | 15.50t/s | 5 | |
| 3 |
| THUDM/GLM-4-9B-0414 |
0.28 s Best: 0.26Worst: 0.32 |
89.85t/s |
| 5 |
| 4 | glm-4-flash-250414 | 0.41 s Best: 0.24Worst: 0.69 | 39.15t/s | 5 |
| 5 | openai/gpt-oss-120b | 0.46 s Best: 0.40Worst: 0.55 | 212.52t/s | 5 |
| 6 | institute-of-science-tokyo/llama-3.1-swallow-70b-instruct-v0.1 | 0.50 s Best: 0.45Worst: 0.62 | 19.06t/s | 5 |
| 7 | glm-4-flash | 0.51 s Best: 0.42Worst: 0.56 | 27.97t/s | 5 |
| 8 | 英伟达/mistralai/devstral-2-123b-instruct-2512 | 0.55 s Best: 0.49Worst: 0.68 | 58.22t/s | 5 |
| 9 | llama3.1-8b | 0.58 s Best: 0.53Worst: 0.66 | 2286.66t/s | 10 |
| 10 | moonshotai/kimi-k2-instruct | 0.60 s Best: 0.56Worst: 0.63 | 24.50t/s | 5 |
| 11 | Qwen/Qwen2.5-7B-Instruct | 0.61 s Best: 0.22Worst: 1.16 | 106.01t/s | 5 |
| 12 | qwen-3-235b-a22b-instruct-2507 | 0.63 s Best: 0.55Worst: 0.75 | 910.42t/s | 5 |
| 13 | Qwen/Qwen2.5-72B-Instruct | 0.65 s Best: 0.37Worst: 1.23 | 32.99t/s | 5 |
| 14 | zhipu/glm-4v-flash | 0.72 s Best: 0.32Worst: 2.82 | 47.32t/s | 10 |
| 15 | openai/gpt-oss-20b | 0.77 s Best: -Worst: 9.67 | 9.08t/s | 30 |
| 16 | zhipu/glm-4-flash | 0.79 s Best: 0.46Worst: 2.64 | 28.24t/s | 555 |
| 17 | mimo-v2-flash | 0.83 s Best: 0.41Worst: 1.15 | 110.54t/s | 5 |
| 18 | kimi-k2.5 | 0.88 s Best: 0.67Worst: 1.24 | 43.53t/s | 5 |
| 19 | qwen-vl-plus-2025-05-07 | 0.88 s Best: 0.52Worst: 2.22 | 99.19t/s | 5 |
| 20 | qwen/qwen2.5-7b | 0.94 s Best: 0.33Worst: 4.18 | 87.73t/s | 40 |
| 21 | gpt-5.4 | 1.10 s Best: 0.30Worst: 2.40 | 25.05t/s | 20 |
| 22 | claude-sonnet-4-6 | 1.12 s Best: 0.82Worst: 1.50 | 43.57t/s | 5 |
| 23 | Qwen/Qwen3-VL-8B-Instruct | 1.20 s Best: 0.35Worst: 3.86 | 70.26t/s | 5 |
| 24 | kimi-k2.5 | 1.29 s Best: 0.97Worst: 1.91 | 1231.67t/s | 5 |
| 25 | deepseek-ai/DeepSeek-V3 | 1.45 s Best: 1.07Worst: 2.17 | 18.81t/s | 5 |
| 26 | MiniMax-M2.7-highspeed | 1.56 s Best: 0.75Worst: 3.12 | 48.72t/s | 5 |
| 27 | gpt-5.2-codex | 1.76 s Best: 1.49Worst: 2.18 | 90.69t/s | 5 |
| 28 | claude-sonnet-4-6 | 1.83 s Best: 1.36Worst: 2.59 | 85.69t/s | 10 |
| 29 | deepseek-v3.2 | 1.86 s Best: 1.34Worst: 2.72 | 23.88t/s | 5 |
| 30 | gpt-5.4 | 1.87 s Best: 1.31Worst: 2.70 | 54.33t/s | 5 |
| 31 | mimo-v2-pro | 1.98 s Best: 0.94Worst: 4.46 | 64.80t/s | 5 |
| 32 | deepseek-chat | 2.10 s Best: 1.70Worst: 2.28 | 35.89t/s | 5 |
| 33 | gpt-5.4(high) | 2.17 s Best: 1.13Worst: 4.93 | 52.01t/s | 10 |
| 34 | gpt-5.4 | 2.17 s Best: 1.76Worst: 3.12 | 50.94t/s | 5 |
| 35 | claude-opus-4-6 | 2.19 s Best: 1.69Worst: 2.83 | 112.88t/s | 5 |
| 36 | moonshot-v1-32k | 2.24 s Best: 1.70Worst: 3.00 | 10.88t/s | 5 |
| 37 | [按量]anti/claude-opus-4-6 | 2.34 s Best: 1.98Worst: 3.13 | 46.66t/s | 5 |
| 38 | qwen/qwen3.5-122b-a10b | 2.35 s Best: 0.29Worst: 12.66 | 37.39t/s | 10 |
| 39 | 酒馆-Flash-New | 2.38 s Best: 1.53Worst: 3.03 | 101.00t/s | 5 |
| 40 | deepseek-ai/DeepSeek-R1-Distill-Qwen-7B | 2.42 s Best: 0.43Worst: 9.62 | 79.34t/s | 5 |
| 41 | 酒馆-Flash | 2.54 s Best: 1.36Worst: 4.49 | 55.55t/s | 5 |
| 42 | Translation | 2.56 s Best: 1.87Worst: 3.24 | 378.97t/s | 5 |
| 43 | anthropic/claude-sonnet-4.6 | 2.59 s Best: 2.54Worst: 2.64 | 17475.88t/s | 5 |
| 44 | deepseek-ai/DeepSeek-V3.2 | 2.61 s Best: 0.71Worst: 8.64 | 23.40t/s | 5 |
| 45 | gemini-3-flash | 2.61 s Best: 2.54Worst: 2.66 | 18164.29t/s | 5 |
| 46 | google/gemini-3-flash | 2.62 s Best: 2.55Worst: 2.65 | 16435.04t/s | 5 |
| 47 | PaddlePaddle/PaddleOCR-VL-1.5 | 3.03 s Best: 0.25Worst: 10.27 | 313.91t/s | 5 |
| 48 | grok-4.20-0309-reasoning | 3.13 s Best: 1.33Worst: 5.74 | 225.06t/s | 5 |
| 49 | stepfun-ai/Step-3.5-Flash | 3.37 s Best: 2.54Worst: 4.14 | 84.22t/s | 5 |
| 50 | gpt-5.3-codex | 3.50 s Best: 2.98Worst: 3.97 | 61.35t/s | 5 |