Inference
EN
Vals AI

SRE Bench

Can AI agents reverse engineer real-world, contamination-free binaries?

ModelsModels with a result.36
Top resultBest result on this benchmark.56.9%GPT-6 Astra
Top-3 spreadPoints from first to third.12.6 pts
UpdatedDate the source changed the results.5 Nov 2026

Result and cost

36 of 36 results

Results

36 results
#ModelResultThe score from the source.CostUS dollars to run one task.TimeTime to run one task.
01GPT-6 Astra56.9% ±3.1$10.2418 min
02GPT-6.1 Sol50.8% ±3.1$2.6933 min
03Gemini 4 Argon44.3% ±3.1$30.451.6 h
04Claude Opus 5.533.6% ±2.9$26.481.6 h
05GPT Daybreak Blue31.7% ±2.9$54.342.9 h
06GPT-5.6 Sol30.5% ±2.9$42.5288 min
06GPT-6 Sol30.5% ±2.9$20.6563 min
08Claude Sonnet 5.530.2% ±2.8$17.271.9 h
09Claude Fable 5.122.9% ±2.6$32.232.0 h
10Claude Opus 5.5 Range21.8% ±2.6$33.133.4 h
11Claude Opus 512.2% ±2.0$23.6381 min
12Muse Wonder Full11.8% ±2.0$33.815.6 h
13Gemini 3.8 Flash11.1% ±1.9$27.462.5 h
14GPT-5.6 Terra9.5% ±1.8$30.863.3 h
15GLM-5.39.2% ±1.8$27.472.8 h
16Claude Haiku 5.55.7% ±1.4$3.1458 min
17Gemini 3.7 Flash4.6% ±1.3$21.471.8 h
18MiMo-V2.6-Flash4.2% ±1.2$0.3953.4 h
19Grok 4.73.8% ±1.2$64.771.7 h
19GPT-5.53.8% ±1.2$17.7746 min
19Muse Spark 1.3 Max3.8% ±1.2$25.861.8 h
22MiMo-V2.6-Pro3.1% ±1.1$0.5332.4 h
23GPT-6 Luna2.7% ±1.0$1.5581 min
24Hy4 preview2.3% ±0.9$4.901.7 h
25Inkling 2 Small1.9% ±0.8$6.672.3 h
25DeepSeek V4 Pro 08131.9% ±0.8$5.582.3 h
27GPT-5.6 Luna1.5% ±0.8$3.441.7 h
28DeepSeek V4.1 Flash 10170.8% ±0.5$0.54961 min
28Grok 4.50.8% ±0.5$13.4266 min
28GLM-5.3-Flash0.8% ±0.5$2.293.2 h
31Step 5 Preview0.4% ±0.4$3.071.6 h
31Claude Sonnet 50.4% ±0.4$18.493.9 h
33Muse Spark 1.20.0%$23.431.7 h
33Gemini 3.1 Pro Preview0.0%$11.3463 min
33GLM-5.20.0%$26.582.5 h
33Inkling0.0%$2.2329 min