InferenceBench: leaderboard
Benchmark for coding agents optimizing inference workloads. Agents tune serving configurations and implementation choices across latency, throughput, and all-in-one scenarios.
Metric: Speedup Score. Source: inferencebench.ai. Status: years away from saturation. 32 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 5 | 8.9 |
| 2 | Claude Fable 5 (Low) | 8.74 |
| 3 | Claude Opus 4.7 | 8.53 |
| 4 | Claude Opus 4.8 | 7.6 |
| 5 | Claude Fable 5 | 7.52 |
| 6 | Claude Opus 4.8 (xHigh) | 7.34 |
| 7 | GLM-5.2 (Max) | 7 |
| 8 | Grok 4.6 | 6.65 |
| 9 | Claude Sonnet 5 | 6.43 |
| 10 | Kimi K2.7 Code | 6.27 |
| 11 | GPT-5.4 (High) | 6.16 |
| 12 | Kimi K3 | 5.7 |
| 13 | Claude Sonnet 4.6 | 5.56 |
| 14 | GPT-5.3 Codex (High) | 5.49 |
| 15 | GPT-5.5 (xHigh) | 5.45 |
Interactive version: theaggregate.ai/benchmark?slug=inferencebench · How It Works · Data refreshed daily, snapshot 2026-09-05.