跳过主要内容
评测来源/Terminal-Bench 4.0(Vals 复跑)
参与排名编程

Terminal-Bench 4.0(Vals 复跑)

Vals AI · Terminal-Bench 4.0(Vals 复跑) 的公开模型评测结果。

官方评测
来源说明

该来源测量模型在「编程」领域的具体交付表现,原始成绩会和统一能力读数分开展示。

限制: 不同版本、推理设置和测试集可能导致分数变化,请结合数据时间阅读。

使用方式

参与排名的来源会影响能力分;参考、交叉参考和观察来源仅用于对比与核验。

本来源的模型成绩

Claude Opus 5.565.1579.62026-11-05
Claude Sonnet 5.564.1478.72026-11-05
GPT-6 Astra59.675.12026-11-05
Claude Fable 5.158.0873.92026-11-05
Gemini 4 Argon57.5873.52026-11-05
GPT-6.1 Sol55.0571.42026-11-05
Claude Opus 553.5370.22026-11-05
GPT-6.1 Terra49.4967.02026-11-05
GPT-6 Sol44.4462.92026-11-05
Claude Fable 541.4160.52026-11-05
GLM 5.338.8958.42026-11-05
GPT-5.6 Sol37.8857.62026-11-05
GLM-5.3 Flash37.3757.22026-11-05
GLM-5.3 Flash37.3757.22026-10-18
GLM-5.3 Flash37.3757.22026-10-17
Claude Haiku 5.536.3656.42026-11-05
Qwen3.8 Max34.3454.82026-11-05
DeepSeek V4.1 Flash 101731.8252.72026-11-05
Step 5 Preview31.8252.72026-11-05
MiMo V2.6 Pro31.3152.32026-11-05
Grok 4.728.7950.32026-11-05
Mistral Large 426.2648.32026-11-05
MiniMax M3.1 Flash25.2547.52026-11-05
Muse Spark 1.324.7547.02026-11-05
MiMo V2.6 Flash24.2446.62026-10-18
MiMo V2.6 Flash24.2446.62026-11-05
MiMo V2.6 Flash24.2446.62026-10-17
Claude Opus 4.823.2345.82026-11-05
GPT-5.6 Terra22.7345.42026-11-05
Gemini 3.8 Flash19.743.02026-11-05
DeepSeek V4.1 Flash19.743.02026-11-05
DeepSeek V4 Flash 073118.6942.22026-11-05
Kimi K317.1741.02026-11-05
Grok 4.617.1741.02026-11-05
DeepSeek V4 Pro 081314.1438.52026-11-05
GPT-6 Luna13.6438.12026-11-05
GPT-5.6 Luna11.6236.52026-11-05
Claude Sonnet 510.6135.72026-11-05
Grok 4.58.5934.02026-11-05
Muse Spark 1.26.0632.02026-11-05
GPT-5.4 mini2.5229.22026-11-05
MiniMax-M31.0127.92026-11-05
MiMo-V2.5-Pro0.5127.52026-11-05
Inkling0.5127.52026-11-05
Inkling Small027.12026-11-05

常见问题