MODELS · BENCHMARKS

AI model rankings

Showing public results from SRE-Bench and AgentRE-Bench V3, ranked by each benchmark’s score. Bar lengths are comparable within each ranking only.

SRE-Bench

Ranked by 成绩 (maximum 100%) · Error bars show source-reported stderr

#Model成绩

Ranked by 总分 (maximum 2)

#Model总分

MODELS · BENCHMARKS

Showing public results from SRE-Bench and AgentRE-Bench V3, ranked by each benchmark’s score. Bar lengths are comparable within each ranking only.

Ranked by 成绩 (maximum 100%) · Error bars show source-reported stderr

#Model成绩

Ranked by 总分 (maximum 2)

#Model总分

3GPT-5.6 SolOpenAI30.53%±2.85

4Claude Fable 5.1Anthropic22.90%±2.60

5Claude Opus 5Anthropic12.21%±2.03

6Gemini 3.7 FlashGoogle4.58%±1.29

7GPT-5.5OpenAI3.82%±1.19

8Hy4Tencent2.29%±0.93

9DeepSeek V4.1 FlashDeepSeek0.76%±0.54

10Grok 4.5Spacexai0.76%±0.54

11GLM-5.2Z.ai0.00%±0.00

DeepSeek V4 FlashDeepSeek

1.1906

5Gemini 3.6 FlashGoogle1.1052

—Claude Fable 5AnthropicNo score reported

Columbia · Pasta Lab · Vals 当期 · Data through 2026-09-23 · Source Vals AI · SRE Bench（经 PublicAI Index）

agentrebench · version3 · Fetched 2026-09-24 · Source GitHub · agentrebench/AgentRE-Bench