Provider Snapshot
9
48.66
554.44
May 11, 2026
Key Takeaways
9 together models are actively benchmarked with 1372 total measurements across 1330 benchmark runs.
Qwen3.5-397B-A17B leads the fleet with 77.80 tokens/second, while deepseek-r1-0528 delivers 40.90 tok/s.
Performance varies by 90.2% across the together model lineup, indicating diverse optimization strategies for different use cases.
Avg time to first token across the fleet is 554.44 ms, showing good responsiveness for interactive applications.
The together model fleet shows varied performance characteristics (41.6% variation coefficient), reflecting diverse model architectures.
Fastest Models
| Provider | Model | Avg Toks/Sec | Min | Max | Avg TTF (ms) |
|---|---|---|---|---|---|
| together | Qwen3.5-397B-A17B | 77.80 | 21.40 | 140.00 | 0.00 |
| together | qwen-3.5-9b | 74.30 | 56.30 | 102.00 | 1540.00 |
| together | qwen-2.5-7b | 63.50 | 49.60 | 71.70 | 370.00 |
| together | mixtral-8x7b | 62.80 | 7.95 | 111.00 | 400.00 |
| together | deepseek-r1 | 42.00 | 21.90 | 59.90 | 600.00 |
| together | deepseek-r1-0528 | 40.90 | 9.32 | 62.90 | 600.00 |
All Models
Complete list of all together 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) |
|---|---|---|---|---|---|
| together | Qwen3.5-397B-A17B | 77.80 | 21.40 | 140.00 | 0.00 |
| together | qwen-3.5-9b | 74.30 | 56.30 | 102.00 | 1540.00 |
| together | deepseek-r1 | 42.00 | 21.90 | 59.90 | 600.00 |
| together | deepseek-r1-0528 | 40.90 | 9.32 | 62.90 | 600.00 |
| together | gemma-3n-e4b-it | 21.00 | 2.38 | 42.30 | 620.00 |
| together | llama-3.3-70b | 31.20 | 6.24 | 55.30 | 860.00 |
| together | mixtral-8x7b | 62.80 | 7.95 | 111.00 | 400.00 |
| together | Kimi-K2.5 | 24.40 | 2.19 | 88.40 | 0.00 |
| together | qwen-2.5-7b | 63.50 | 49.60 | 71.70 | 370.00 |
Featured Models
Frequently Asked Questions
Based on recent tests, Qwen3.5-397B-A17B shows the highest average throughput among tracked together models.
This provider summary aggregates 1372 individual prompts measured across 1330 monitoring runs over the past month.