跳过主要内容
评测来源/Terminal-Bench Science(Vals 复跑)
参与排名推理

Terminal-Bench Science(Vals 复跑)

Vals AI · Terminal-Bench Science(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「推理」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

GPT-6 Astra62.8689.42026-11-05
GPT-6.1 Sol52.8680.92026-11-05
Claude Opus 5.547.1476.12026-11-05
Claude Sonnet 5.547.1476.12026-11-05
Gemini 4 Argon44.2973.72026-11-05
Claude Fable 5.14070.02026-11-05
GPT-6 Sol3061.62026-11-05
Claude Opus 527.1459.12026-11-05
GPT-5.6 Sol2053.12026-11-05
Claude Fable 515.7149.42026-11-05
Qwen3.8 Max12.8647.02026-11-05
Grok 4.611.4345.82026-11-05
GPT-5.6 Terra1044.62026-11-05
Muse Spark 1.31044.62026-11-05
Grok 4.71044.62026-11-05
Claude Haiku 5.51044.62026-11-05
Gemini 3.8 Flash8.5743.42026-11-05
Claude Sonnet 55.7141.02026-11-05
GLM 5.35.7141.02026-11-05
MiMo V2.6 Flash4.2939.82026-11-05
Claude Opus 4.84.2939.82026-11-05
Gemini 3.6 Flash4.2939.82026-11-05
DeepSeek V4 Flash 07314.2939.82026-11-05
DeepSeek V4 Pro 08134.2939.82026-11-05
GPT-6 Luna4.2939.82026-11-05
GLM-5.3 Flash2.8638.62026-11-05
MiMo V2.6 Pro2.8638.62026-11-05
Step 5 Preview2.8638.62026-11-05
Kimi K31.4337.32026-11-05
GPT-5.6 Luna036.12026-11-05
DeepSeek V4.1 Flash036.12026-11-05

常见问题