AccelerateBench v1.1 - Speedup >= 1.25x: leaderboard

Share of 32 AccelerateBench v1.1 HPC program optimization tasks (arXiv:2611.20882v1, Tables 1-2) where the agent's final program passes correctness checks and achieves at least a 1.25x speedup (20% runtime reduction).

Metric: Tasks correct and >=1.25x faster (%). Source: arxiv.org. Saturation forecast: Around February 2027. 16 models tracked. Updated 21 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Claude Sonnet 5.5High100#24
2Claude Opus 5.5High84.4#9
3GPT-6.1 SolHigh78.1#15
4GPT-6 AstraHigh75#10
5GPT-6.1 TerraHigh75#25
6hy4High75#40
7Hy4 previewHigh71.9#74
8Kimi K3High68.8#61
9GLM-5.3 FlashHigh68.8#133
10DeepSeek V4.1 Flash (1017)High65.6#47
11GLM-5.3High65.6#72
12DeepSeek V4 ProHigh62.5#196
13DeepSeek V4 FlashHigh56.2#267
14Grok 4.6High50#50
15GLM-5.2High46.9#150

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=acceleratebench-v1-1-speedup-over-1-25x · How It Works · Data refreshed daily, snapshot 2026-11-21.