Provider Snapshot
10
67.78
632.00
Mar 8, 2026
Key Takeaways
10 together models are actively benchmarked with 1879 total measurements across 1736 benchmark runs.
llama-3.1-8b leads the fleet with 142.00 tokens/second, while llama-3.2-3b delivers 57.90 tok/s.
Performance varies by 145.3% across the together model lineup, indicating diverse optimization strategies for different use cases.
Avg time to first token across the fleet is 632.00 ms, showing good responsiveness for interactive applications.
The together model fleet shows varied performance characteristics (44.5% variation coefficient), reflecting diverse model architectures.
Fastest Models
| Provider | Model | Avg Toks/Sec | Min | Max | Avg TTF (ms) |
|---|---|---|---|---|---|
| together | llama-3.1-8b | 142.00 | 3.05 | 232.00 | 380.00 |
| together | qwen-2.5-7b | 94.20 | 11.90 | 145.00 | 240.00 |
| together | llama-3.1-70b | 72.60 | 7.16 | 129.00 | 390.00 |
| together | mistral-7b | 70.70 | 2.11 | 90.80 | 540.00 |
| together | mixtral-8x7b | 60.80 | 24.80 | 114.00 | 150.00 |
| together | llama-3.2-3b | 57.90 | 5.45 | 121.00 | 1480.00 |
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 | llama-3.3-70b | 53.90 | 1.29 | 146.00 | 1350.00 |
| together | deepseek-r1 | 45.50 | 1.34 | 113.00 | 1000.00 |
| together | mistral-7b | 70.70 | 2.11 | 90.80 | 540.00 |
| together | qwen-2.5-72b | 55.70 | 49.40 | 62.10 | 340.00 |
| together | qwen-2.5-7b | 94.20 | 11.90 | 145.00 | 240.00 |
| together | mixtral-8x7b | 60.80 | 24.80 | 114.00 | 150.00 |
| together | llama-3.2-3b | 57.90 | 5.45 | 121.00 | 1480.00 |
| together | llama-3.1-405b | 24.50 | 24.50 | 24.50 | 450.00 |
| together | llama-3.1-70b | 72.60 | 7.16 | 129.00 | 390.00 |
| together | llama-3.1-8b | 142.00 | 3.05 | 232.00 | 380.00 |
Featured Models
Frequently Asked Questions
Based on recent tests, llama-3.1-8b shows the highest average throughput among tracked together models.
This provider summary aggregates 1879 individual prompts measured across 1736 monitoring runs over the past month.