Provider Snapshot
9
30.37
1.22
Oct 18, 2025
Key Takeaways
9 anthropic models are actively benchmarked with 10194 total measurements across 10034 benchmark runs.
claude-3-haiku leads the fleet with 87.10 tokens/second, while claude-4-sonnet delivers 20.90 tok/s.
Performance varies by 316.7% across the anthropic model lineup, indicating diverse optimization strategies for different use cases.
Median time to first token across the fleet is 1.22 ms, showing excellent responsiveness for interactive applications.
The anthropic model fleet shows varied performance characteristics (69.1% variation coefficient), reflecting diverse model architectures.
Fastest Models
| Provider | Model | Avg Toks/Sec | Min | Max | Avg TTF (ms) |
|---|---|---|---|---|---|
| anthropic | claude-3-haiku | 87.10 | 16.00 | 108.00 | 0.32 |
| anthropic | claude-3-5-sonnet | 35.50 | 6.17 | 55.40 | 0.80 |
| anthropic | claude-3-5-haiku | 30.70 | 10.60 | 41.40 | 0.50 |
| anthropic | claude-3-7-sonnet | 26.30 | 1.05 | 42.40 | 1.26 |
| anthropic | claude-3-opus | 22.10 | 3.00 | 27.00 | 0.83 |
| anthropic | claude-4-sonnet | 20.90 | 1.51 | 33.50 | 1.46 |
All Models
Complete list of all anthropic 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) |
|---|---|---|---|---|---|
| anthropic | claude-3-haiku | 87.10 | 16.00 | 108.00 | 0.32 |
| anthropic | claude-3-5-haiku | 30.70 | 10.60 | 41.40 | 0.50 |
| anthropic | claude-3-5-sonnet | 35.50 | 6.17 | 55.40 | 0.80 |
| anthropic | claude-3-7-sonnet | 26.30 | 1.05 | 42.40 | 1.26 |
| anthropic | claude-3-opus | 22.10 | 3.00 | 27.00 | 0.83 |
| anthropic | claude-4-opus | 16.80 | 1.00 | 25.00 | 1.46 |
| anthropic | claude-4-sonnet | 20.90 | 1.51 | 33.50 | 1.46 |
| anthropic | Claude Opus 4.1 | 16.10 | 1.01 | 25.30 | 2.44 |
| anthropic | claude-sonnet-4.5 | 17.80 | 1.27 | 22.30 | 1.91 |
Featured Models
Frequently Asked Questions
Based on recent tests, claude-3-haiku shows the highest average throughput among tracked anthropic models.
This provider summary aggregates 10194 individual prompts measured across 10034 monitoring runs over the past month.