Cloud LLM benchmarks
- Models
- 244
- Providers
- 9
- Samples
- 67,321
- Median tok/s
- 33
- p90 tok/s
- 74
- Max tok/s
- 295
- Median spread
- 143%
- Median TTFT
- 1.37s
Throughput distribution · tok/s
- GPT-oss-safeguard-20b groq295 tok/s
- llama-3.1-8b groq221 tok/s
- Qwen3.6-27B groq216 tok/s
- qwen-3-32b groq174 tok/s
- Google: Nano Banana (Gemini 2.5 Flash Image) google159 tok/s
- llama-3.3-70b groq150 tok/s
- gemma-4-31b cerebras138 tok/s
- nova-micro bedrock121 tok/s
- llama-4-scout groq117 tok/s
- LFM2.5-8B-A1B together110 tok/s
- qwen-3.5-35b-a3b deepinfra110 tok/s
- llama-4-maverick bedrock107 tok/s
- qwen-2-1.5b-instruct together105 tok/s
- GPT-oss-120b fireworks97 tok/s
- llama-3.1-8b bedrock96 tok/s
- llama-4-scout bedrock95 tok/s
- nova-lite bedrock91 tok/s
- nova-pro bedrock91 tok/s
- GPT-5.1-codex-mini openai85 tok/s
- llama-3.3-70b bedrock84 tok/s
- GPT-5 Nano openai84 tok/s
- mistral-7b bedrock82 tok/s
Visible-token throughput, same basis as the table · curves normalised per model · 1 model's tail runs past the axis · 222 slower models not drawn, all 244 are in the table below
By provider
| Provider | Models | Median | p90 | Best | Spread | TTFT |
|---|---|---|---|---|---|---|
| groq | 6 | 174 | 295 | 295 | 146% | — |
| cerebras | 1 | 138 | 138 | 138 | 148% | 0.80 |
| together | 20 | 54 | 79 | 110 | 189% | — |
| bedrock | 22 | 51 | 96 | 121 | 111% | 0.38 |
| fireworks | 20 | 39 | 65 | 97 | 180% | — |
| openai | 38 | 38 | 61 | 85 | 154% | 1.88 |
| deepinfra | 123 | 27 | 48 | 110 | 146% | — |
| anthropic | 10 | 24 | 29 | 43 | 113% | 1.37 |
| 4 | 2 | 159 | 159 | 107% | 0.92 |
Throughput × spread · 244 models
Full results
244 of 244 modelsThroughput over time · shared scale
Method. A cron job calls each model's live API endpoint on a schedule and records what came back. Mean, min and max are over completed samples in the selected window, and n is how many there were. A missing value means the endpoint returned an error or the model was not yet in the catalogue.
Spread is (max − min) ÷ mean, so it measures run-to-run variation rather than absolute speed. TTFT is seconds to the first visible token; runs that emit only reasoning tokens are left out of that average. Mean counts visible output tokens where a provider reports them and falls back to generated throughput where it does not, which is why Gen is higher for models that think before answering.