Gemini 3.1 Pro · τ²-bench(Telecom section) 99.3% · 附件证据(eval-tau-telecom-launch)

Gemini 3.1 Pro 的 τ²-bench · Telecom section 官方出处。Gemini 3.1 Pro Thinking (High) · 99.3%。见《Gemini 3.1 Pro Model Evaluation — launch-day version》PDF p.3, results table, tau2-bench / Telecom row, Gemini 3.1 Pro Thinking (High) column。证据 eval-tau-telecom-launch。

原文与定位

Gemini 3.1 Pro · τ²-bench(Telecom section),Gemini 3.1 Pro Thinking (High)配置,Success rate 99.3%(具体成绩公布时间未确认)。见《Gemini 3.1 Pro Model Evaluation — launch-day version》PDF p.3, results table, tau2-bench / Telecom row, Gemini 3.1 Pro Thinking (High) column。 查看成绩条件

tau2-bench / Telecom / Gemini 3.1 Pro Thinking (High): 99.3%
官方材料
Gemini 3.1 Pro Model Evaluation — launch-day version · PDF 第 3 页
定位
附件 · PDF p.3, results table, tau2-bench / Telecom row, Gemini 3.1 Pro Thinking (High) column
材料版本
3-page evaluation PDF archived 2026-02-19 16:22:46 UTC; separate from current methodology revision.
版本公布
未确认
披露时间
· 已有首发依据
采集时间
历史依据
已实际下载官方PDF首发当天回放 https://web.archive.org/web/20260219162246id_/https://storage.googleapis.com/deepmind-media/gemini/gemini_3-1_pro_model_evaluation.pdf;Memento-Datetime Thu, 19 Feb 2026 16:22:46 GMT。3页全文已核读,p.3主表实际渲染核查;与当前版本的差异为SWE-bench Pro方法新增based on the public set and字样,原始分数未变。
引用文本

Gemini 3.1 Pro · τ²-bench(Telecom section),Gemini 3.1 Pro Thinking (High)配置,Success rate 99.3%(具体成绩公布时间未确认)。官方出处 Gemini 3.1 Pro Model Evaluation — launch-day version(附件,3-page evaluation PDF archived 2026-02-19 16:22:46 UTC; separate from current methodology revision.)PDF p.3, results table, tau2-bench / Telecom row, Gemini 3.1 Pro Thinking (High) column,https://web.archive.org/web/20260219162246id_/https://storage.googleapis.com/deepmind-media/gemini/gemini_3-1_pro_model_evaluation.pdf#page=3。Benchlog https://benchlog.lekshome.cc/evidence/gemini-3-1-pro/eval-tau-telecom-launch/

关联成绩与条件

Gemini 3.1 Pro · τ²-bench 成绩与配置
指标与成绩配置与评测条件证据关系
Success rate99.3%原文 99.3% · 单位 %
公布 未确认
Gemini 3.1 Pro Thinking (High)
设置
Gemini API model id gemini-3.1-pro-preview; default sampling settings; evaluation table column specifies Thinking (High). All scores are pass@1 except where noted.
未知项
Exact thinking budget and sampling parameters are not listed.
工具
工具条件未记录
推理预算
high
框架
Sierra framework
协议
Telecom category; standard Sierra framework; prompt adjustment for environment instructions; user model uses Gemini with a system instruction.
条件
Airline category excluded from the report due to lower-quality grading.

直接支持该成绩

在排名中查看此成绩