Provider Snapshot
39
41.02
468.72
May 20, 2026
Key Takeaways
39 deepinfra models are actively benchmarked with 5539 total measurements across 4960 benchmark runs.
qwen-3.5-2b leads the fleet with 200.00 tokens/second, while qwen-3.5-397b-a17b delivers 62.50 tok/s.
Performance varies by 220.0% across the deepinfra model lineup, indicating diverse optimization strategies for different use cases.
Avg time to first token across the fleet is 468.72 ms, showing excellent responsiveness for interactive applications.
The deepinfra model fleet shows varied performance characteristics (95.3% variation coefficient), reflecting diverse model architectures.
Fastest Models
| Provider | Model | Avg Toks/Sec | Min | Max | Avg TTF (ms) |
|---|---|---|---|---|---|
| deepinfra | qwen-3.5-2b | 200.00 | 44.00 | 231.00 | 1320.00 |
| deepinfra | qwen-3.5-0.8b | 142.00 | 32.40 | 235.00 | 420.00 |
| deepinfra | qwen-3.5-4b | 141.00 | 35.00 | 174.00 | 380.00 |
| deepinfra | qwen-3.5-35b-a3b | 79.70 | 7.96 | 161.00 | 340.00 |
| deepinfra | qwen-3.5-122b-a10b | 79.20 | 1.63 | 115.00 | 260.00 |
| deepinfra | qwen-3.5-397b-a17b | 62.50 | 4.15 | 103.00 | 290.00 |
All Models
Complete list of all deepinfra models tracked in the benchmark system. Click any model name to view detailed performance data.
| Provider | Model | Avg Toks/Sec | Min | Max | Avg TTF (ms) |
|---|---|---|---|---|---|
| deepinfra | Seed-2.0-pro | 2.17 | 1.03 | 3.02 | 0.00 |
| deepinfra | qwen-2.5-72b | 17.90 | 1.25 | 33.50 | 3680.00 |
| deepinfra | Qwen 2.5 Coder 32B | 45.80 | 11.00 | 67.90 | 470.00 |
| deepinfra | qwen-3-14b | 34.60 | 3.69 | 49.50 | 680.00 |
| deepinfra | qwen-3-235b | 12.60 | 2.22 | 30.00 | 2010.00 |
| deepinfra | Qwen3-235B-A22B-Thinking-2507 | 6.67 | 0.83 | 17.40 | 0.00 |
| deepinfra | Qwen3-Coder-480B-A35B-Instruct-Turbo | 26.10 | 3.93 | 78.70 | 0.00 |
| deepinfra | Qwen3-Max | 17.50 | 8.29 | 20.80 | 0.00 |
| deepinfra | Qwen3-Max-Thinking | 16.70 | 10.10 | 20.20 | 0.00 |
| deepinfra | Qwen3-Next-80B-A3B-Instruct | 50.30 | 4.14 | 115.00 | 0.00 |
| deepinfra | qwen-3.5-0.8b | 142.00 | 32.40 | 235.00 | 420.00 |
| deepinfra | qwen-3.5-122b-a10b | 79.20 | 1.63 | 115.00 | 260.00 |
| deepinfra | qwen-3.5-27b | 42.90 | 11.20 | 69.00 | 300.00 |
| deepinfra | qwen-3.5-2b | 200.00 | 44.00 | 231.00 | 1320.00 |
| deepinfra | qwen-3.5-35b-a3b | 79.70 | 7.96 | 161.00 | 340.00 |
| deepinfra | qwen-3.5-397b-a17b | 62.50 | 4.15 | 103.00 | 290.00 |
| deepinfra | qwen-3.5-4b | 141.00 | 35.00 | 174.00 | 380.00 |
| deepinfra | Qwen3.6-27B | 39.80 | 4.47 | 75.90 | 0.00 |
| deepinfra | DeepSeek-R1-0528 | 29.50 | 9.81 | 56.50 | 0.00 |
| deepinfra | deepseek-v3.2 | 10.40 | 4.14 | 18.30 | 0.00 |
| deepinfra | DeepSeek-V4-Pro | 21.90 | 1.60 | 49.60 | 0.00 |
| deepinfra | gemini-2.5-flash | 34.90 | 9.14 | 47.90 | 0.00 |
| deepinfra | gemma-3-27b-it | 20.40 | 9.61 | 37.50 | 460.00 |
| deepinfra | gemma-3-4b-it | 35.00 | 4.50 | 56.40 | 490.00 |
| deepinfra | llama-2-70b | 23.90 | 2.53 | 37.90 | 530.00 |
| deepinfra | llama-3.2-11b | 37.90 | 16.50 | 58.20 | 440.00 |
| deepinfra | llama-3.2-1b | 36.10 | 20.30 | 48.30 | 350.00 |
| deepinfra | llama-3.2-3b | 36.30 | 24.70 | 58.90 | 350.00 |
| deepinfra | llama-3.2-90b | 24.40 | 7.05 | 44.30 | 390.00 |
| deepinfra | llama-3.3-70b | 18.00 | 2.05 | 32.90 | 830.00 |
| deepinfra | llama-3-70b | 23.60 | 5.12 | 38.00 | 520.00 |
| deepinfra | llama-3-8b | 37.60 | 18.80 | 62.00 | 330.00 |
| deepinfra | llama-3.1-405b | 15.30 | 2.34 | 24.70 | 1060.00 |
| deepinfra | llama-3.1-70b | 21.60 | 5.30 | 39.40 | 590.00 |
| deepinfra | llama-3.1-8b | 31.60 | 3.51 | 54.40 | 550.00 |
| deepinfra | devstral-small | 36.10 | 4.50 | 61.50 | 700.00 |
| deepinfra | mistral-7b | 36.60 | 11.90 | 60.60 | 540.00 |
| deepinfra | Kimi-K2.6 | 18.20 | 2.69 | 54.00 | 0.00 |
| deepinfra | Llama-3.3-Nemotron-Super-49B-v1.5 | 33.10 | 4.65 | 50.70 | 0.00 |
Featured Models
Frequently Asked Questions
Based on recent tests, qwen-3.5-2b shows the highest average throughput among tracked deepinfra models.
This provider summary aggregates 5539 individual prompts measured across 4960 monitoring runs over the past month.