Speed AI model rankings
The Speed track ranks 40 AI models by median output tokens per second from production API traffic. Snapshot 2026-10-14, updated weekly.
Snapshot 2026-10-14 · updated weekly
| Rank | Model | Vendor | Speed | GPQA | Context | Price (in / out) | Speed |
|---|---|---|---|---|---|---|---|
1— | Celeris-1 | Celeris | 1,431 | 42.6% | 128K | $0.20 / $0.80 | 1431 |
2— | Mercury 2.5 | Inception Labs | 583 | 48.2% | 128K | $0.10 / $0.40 | 583 |
3— | Qwen3.8 Flash | Alibaba | 412 | 66.8% | 256K | $0.06 / $0.25 | 412 |
4NEW | Laya 1 | Laya Research | 368 | 38.4% | 64K | $0.05 / $0.20 | 368 |
5— | MiniMax H3 Flash | MiniMax | 358 | 72.6% | 512K | $0.10 / $0.40 | 358 |
6▲2 | Gemini 3.8 Flash | 342 | 78.9% | 1M | $0.15 / $0.60 | 342 | |
7NEW | Jev 1 | Jev Labs | 322 | 59.4% | 128K | $0.08 / $0.30 | 322 |
8— | Kimi K3 Flash | Moonshot AI | 296 | 70.2% | 256K | $0.14 / $0.55 | 296 |
9— | GPT-5.6 Luna | OpenAI | 286 | 74.2% | 512K | $1.20 / $5.50 | 286 |
10NEW | Clef 2 | Clef AI | 276 | 56.9% | 128K | $0.10 / $0.35 | 276 |
11▲3 | GLM-5.3 Flash | Zhipu AI | 268 | 80.4% | 256K | $0.12 / $0.50 | 268 |
12NEW | Strata 2 | Strata Labs | 248 | 62.8% | 128K | $0.12 / $0.45 | 248 |
13NEW | Claude Haiku 5.5 | Anthropic | 232 | 83.2% | 500K | $0.80 / $4 | 232 |
14— | Grok-4 Fast Reasoning | xAI | 210 | 81.5% | 2M | $0.80 / $3.20 | 210 |
15NEW | Pixel Canary | 210 | 31.2% | 32K | Free / Free | 210 | |
16— | Qwen3.8 27B | Alibaba | 186 | 76.8% | 256K | $0.15 / $0.60 | 186 |
17— | Gemma 4 27B | 174 | 67.4% | 256K | $0.18 / $0.70 | 174 | |
18— | Gemini 3.1 Pro | 172 | 79.8% | 1M | $1.25 / $5 | 172 | |
19— | Gemini 3 Pro | 168 | 80.9% | 1M | $1.50 / $6 | 168 | |
20▲2 | Gemini 4 Argon | 165 | 92.1% | 2M | $2.50 / $12 | 165 | |
21— | Gemini 4.0 | 158 | 87.9% | 2M | $2.20 / $10 | 158 | |
22— | Claude Sonnet 5.5 | Anthropic | 156 | 92.8% | 500K | $3 / $15 | 156 |
23— | Claude Fable 5.1 | Anthropic | 142 | 93.4% | 400K | $4 / $20 | 142 |
24— | MiMo-V2.6-Pro | Xiaomi | 142 | 75.4% | 256K | $0.25 / $1 | 142 |
25— | Claude Fable 5 | Anthropic | 138 | 91.9% | 400K | $3.50 / $18 | 138 |
26— | GPT-5.6 Sol | OpenAI | 134 | 88.4% | 512K | $2.80 / $16 | 134 |
27— | Mistral Large 3.5 | Mistral AI | 132 | 80.7% | 256K | $0.80 / $2.40 | 132 |
28▼1 | GPT-6 Astra | OpenAI | 128 | 96% | 1M | $5 / $30 | 128 |
29— | Muse Wonder | Meta | 126 | 85.9% | 256K | $0.45 / $1.60 | 126 |
30▲1 | MiniMax H3 | MiniMax | 121 | 85.8% | 1M | $0.60 / $2.20 | 121 |
31— | GPT-6 Sol | OpenAI | 118 | 90.2% | 800K | $4 / $24 | 118 |
32— | GLM-5.3 | Zhipu AI | 118 | 85.2% | 256K | $0.40 / $1.50 | 118 |
33▲1 | GPT-6.1 Sol | OpenAI | 112 | 94.9% | 1M | $6 / $36 | 112 |
34— | DeepSeek V4.5 | DeepSeek | 112 | 83.7% | 256K | $0.30 / $1.20 | 112 |
35NEW | DeepSeek Harness | DeepSeek | 105 | 86.5% | 256K | $0.35 / $1.40 | 105 |
36▲1 | Qwen3.8 Next | Alibaba | 104 | 82.9% | 512K | $0.50 / $1.80 | 104 |
37— | Kimi K3 | Moonshot AI | 98 | 93.5% | 512K | $0.50 / $2 | 98 |
38▼1 | Claude Opus 5 | Anthropic | 96 | 90.7% | 400K | $6 / $30 | 96 |
39— | Claude Opus 5.5 | Anthropic | 88 | 95.8% | 500K | $8 / $40 | 88 |
40— | Muse Wonder Full | Meta | 88 | 88.1% | 512K | $0.90 / $3.20 | 88 |
Speed FAQ
Which AI model is best for Speed?
Celeris-1 leads the Speed track as of October 2026. The full table above ranks all 40 tracked models on this dimension, with the underlying values shown for every row.
How is the Speed ranking calculated?
This track ranks models by median output tokens per second from production API traffic. Scores are re-verified weekly against public benchmark sources and API rate cards, and the snapshot date is shown on every table so you can see exactly how fresh each value is.