MODEL HISTORY

Devstral 2 指标历史

Devstral 2 的公开评测事实历史,共 90 条观测、66 项指标;只记录来源变化,不插值、不合成本站总分。

数据快照:2026-09-20T01:21:54.000Z查看原始来源 ↗
历史观测90仅在事实变化时追加
指标数量66不同量纲保持分离
后续变更24初始观测不计作变更
首次收录2026/07/21本站可追溯起点
最近观测2026/09/20不是模型发布日期

怎样理解这份历史

每条记录对应来源快照中的一个原始指标。同一口径下相同值重复抓取不会制造新记录;数值、临时状态或来源状态变化时追加,核实后的方法版本另立指标身份。首个快照是本站的可追溯起点,不代表该指标第一次对外发布。当前模型来源为 Artificial Analysis。

Index、Elo、百分比等单位不能互相相减,也不会在这里合成为“综合总分”。

本页含 AA Intelligence Index v4.3.2 的完整方法观测。该版本对 GDPval-AA v2.1 与 AA-Briefcase v1.1 的 Elo 拟合使用 Crowd-BT;旧 v4.3 与 v4.3.1 点位仍保留各自版本、值与日期,不能将跨版本差额解释成模型能力变化。查看当前方法边界 →

本页还含来源记录为裁判协议 v4.3.1 的独立实时 Elo 历史。它是当时的单项榜单协议,不被当前完整 Index v4.3.2 重标;未区分裁判协议的更早 Elo 仅作历史参考。Intelligence Index 中的 Elo 贡献在模型加入时固定,不能用独立实时 Elo 重算指数。核对 AA 方法说明 ↗

APPEND-ONLY FACTS

观测记录

按观测时间倒序;同日多项指标分别保留。

Devstral 2 的公开评测事实历史
观测日期基准 / 版本指标原始值来源状态
Artificial Analysis intelligence component evaluationsv4.3.2AutomationBench-AA partial scorefraction3.1%Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3.2当前完整方法 v4.3.2 · Crowd-BT Elo;不由实时 Elo 重算 Intelligence IndexAA-Briefcase overall Eloelo472Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3.2CritPtfraction0.0%Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3.2GDP.pdf all-passfraction2.4%Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3.2当前完整方法 v4.3.2 · Crowd-BT Elo;不由实时 Elo 重算 Intelligence IndexGDPval-AA v2elo531Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3.2Humanity's Last Examfraction3.6%Artificial Analysis ↗第三方评测有效
Artificial Analysis capability indicesv4.3.2Intelligence Indexindex8.6Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3.2AA-LCR v1.1fraction32.3%Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3.2AA-Omniscienceindex-46.7Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3.2AA-Omniscience accuracyfraction20.8%Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3.2AA-Omniscience hallucination ratefraction85.3%Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3.2SciCodefraction32.8%Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3.2Terminal-Bench v4.0fraction0.0%Artificial Analysis ↗第三方评测有效
Artificial Analysis GDPval-AA任务 v2 · 裁判 v4.3.1历史参考 · 裁判协议 v4.3.1;不与当前完整方法 v4.3.2 跨协议比较Eloelo689Artificial Analysis ↗第三方评测有效
Artificial Analysis AA-Briefcase来源未标任务版本 · 裁判 v4.3.1历史参考 · 裁判协议 v4.3.1;不与当前完整方法 v4.3.2 跨协议比较Eloelo483Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3.1历史参考 · 裁判协议 v4.3.1;不与当前完整方法 v4.3.2 跨协议比较AA-Briefcase overall Eloelo483Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3.1历史参考 · 裁判协议 v4.3.1;不与当前完整方法 v4.3.2 跨协议比较GDPval-AA v2elo689Artificial Analysis ↗第三方评测有效
Artificial Analysis Openness Indexvsource-unversionedModel availabilitypoints5.0Artificial Analysis ↗第三方评测有效
Artificial Analysis Openness Indexvsource-unversionedModel transparencypoints0.0Artificial Analysis ↗第三方评测有效
Artificial Analysis Openness Indexvsource-unversionedOpenness Indexindex27.8Artificial Analysis ↗第三方评测有效
Artificial Analysis Openness Indexvsource-unversionedPosttraining data accesspoints0.0Artificial Analysis ↗第三方评测有效
Artificial Analysis Openness Indexvsource-unversionedPosttraining data licensepoints0.0Artificial Analysis ↗第三方评测有效
Artificial Analysis Openness Indexvsource-unversionedPretraining data accesspoints0.0Artificial Analysis ↗第三方评测有效
Artificial Analysis Openness Indexvsource-unversionedPretraining data licensepoints0.0Artificial Analysis ↗第三方评测有效
Artificial Analysis GDPval-AAv2历史参考 · 原记录未区分当前裁判协议,不作跨协议能力变化比较Eloelo689Artificial Analysis ↗第三方评测有效
Artificial Analysis AA-Briefcasevsource-unversioned历史参考 · 原记录未区分当前裁判协议,不作跨协议能力变化比较Eloelo483Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3AutomationBench-AA partial scorefraction3.1%Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3历史参考 · 原记录未区分当前裁判协议,不作跨协议能力变化比较AA-Briefcase overall Eloelo483Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3CritPtfraction0.0%Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3GDP.pdf all-passfraction2.4%Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3历史参考 · 原记录未区分当前裁判协议,不作跨协议能力变化比较GDPval-AA v2elo689Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3Humanity's Last Examfraction3.6%Artificial Analysis ↗第三方评测有效
Artificial Analysis capability indicesv4.3Intelligence Indexindex9.4Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3AA-LCR v1.1fraction32.3%Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3AA-Omniscienceindex-46.7Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3AA-Omniscience accuracyfraction20.8%Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3AA-Omniscience hallucination ratefraction85.3%Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3SciCodefraction32.8%Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.3Terminal-Bench v4.0fraction0.0%Artificial Analysis ↗第三方评测有效
Artificial Analysis model aggregatevpublic-model-v1端到端响应时间seconds11.6 秒Model Coordinate repository snapshots第三方评测有效
Artificial Analysis model aggregatevpublic-model-v1生成速度tokens_per_second52.0 Token/sModel Coordinate repository snapshots第三方评测有效
Artificial Analysis model aggregatevpublic-model-v1首个回答 Token 等待seconds2.0 秒Model Coordinate repository snapshots第三方评测有效
Artificial Analysis capability indicesvpublic-model-v1Agentic Indexindex10.6Artificial Analysis第三方评测有效
Artificial Analysis model aggregatevpublic-model-v1端到端响应时间seconds11.4 秒Model Coordinate repository snapshots第三方评测有效
Artificial Analysis intelligence component evaluationsv4.1历史参考 · 原记录未区分当前裁判协议,不作跨协议能力变化比较GDPval-AAelo744Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.1GDPval-AA normalizedfraction12.2%Artificial Analysis ↗第三方评测有效
Artificial Analysis capability indicesvpublic-model-v1Intelligence Indexindex19.2Artificial Analysis第三方评测有效
Artificial Analysis intelligence component evaluationsv4.1AA-LCRfraction31.3%Artificial Analysis第三方评测有效
Artificial Analysis intelligence component evaluationsv4.1AA-Omniscienceindex-46.7Artificial Analysis第三方评测有效
Artificial Analysis model aggregatevpublic-model-v1生成速度tokens_per_second51.4 Token/sModel Coordinate repository snapshots第三方评测有效
Artificial Analysis intelligence component evaluationsv4.1Tau Bankingfraction10.5%Artificial Analysis第三方评测有效
Artificial Analysis model aggregatevpublic-model-v1首个回答 Token 等待seconds1.6 秒Model Coordinate repository snapshots第三方评测有效
Artificial Analysis model aggregatevpublic-model-v1端到端响应时间seconds15.3 秒Model Coordinate repository snapshots第三方评测有效
Artificial Analysis model aggregatevpublic-model-v1生成速度tokens_per_second35.6 Token/sModel Coordinate repository snapshots第三方评测有效
Artificial Analysis model aggregatevpublic-model-v1首个回答 Token 等待seconds1.3 秒Model Coordinate repository snapshots第三方评测有效
Artificial Analysis model aggregatevpublic-model-v1端到端响应时间seconds26.5 秒Model Coordinate repository snapshots第三方评测有效
Artificial Analysis intelligence component evaluationsv4.1历史参考 · 原记录未区分当前裁判协议,不作跨协议能力变化比较GDPval-AAelo745Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.1GDPval-AA normalizedfraction12.3%Artificial Analysis ↗第三方评测有效
Artificial Analysis model aggregatevpublic-model-v1生成速度tokens_per_second19.9 Token/sModel Coordinate repository snapshots第三方评测有效
Artificial Analysis model aggregatevpublic-model-v1首个回答 Token 等待seconds1.3 秒Model Coordinate repository snapshots第三方评测有效
Artificial Analysis model aggregatevpublic-model-v1端到端响应时间seconds8.9 秒Model Coordinate repository snapshots第三方评测有效
Artificial Analysis model aggregatevpublic-model-v1生成速度tokens_per_second65.9 Token/sModel Coordinate repository snapshots第三方评测有效
Artificial Analysis model aggregatevpublic-model-v1首个回答 Token 等待seconds1.3 秒Model Coordinate repository snapshots第三方评测有效
Artificial Analysis capability indicesvpublic-model-v1Agentic Indexindex10.6Artificial Analysis第三方评测有效
Artificial Analysis agentic capability indexvunversionedagentic capability indexindex10.6Artificial Analysis第三方评测有效
Artificial Analysis coding capability indexvunversionedcoding capability indexindex31.3Artificial Analysis第三方评测有效
Artificial Analysis finance capability indexvunversionedfinance capability indexindex13.2Artificial Analysis第三方评测有效
Artificial Analysis strategy capability indexvunversionedstrategy capability indexindex15.2Artificial Analysis第三方评测有效
Artificial Analysis legal capability indexvunversionedlegal capability indexindex13.9Artificial Analysis第三方评测有效
Artificial Analysis healthcare capability indexvunversionedhealthcare capability indexindex13.2Artificial Analysis第三方评测有效
Artificial Analysis engineering capability indexvunversionedengineering capability indexindex22.2Artificial Analysis第三方评测有效
Artificial Analysis economics capability indexvunversionedeconomics capability indexindex17.2Artificial Analysis第三方评测有效
Artificial Analysis capability indicesvpublic-model-v1Coding Indexindex31.3Artificial Analysis第三方评测有效
Artificial Analysis model aggregatevpublic-model-v1上下文窗口tokens256,000Model Coordinate repository snapshots第三方评测有效
Artificial Analysis intelligence component evaluationsv4.1CritPtfraction0.0%Artificial Analysis第三方评测有效
Artificial Analysis model aggregatevpublic-model-v1端到端响应时间seconds8.5 秒Model Coordinate repository snapshots第三方评测有效
Artificial Analysis intelligence component evaluationsv4.1历史参考 · 原记录未区分当前裁判协议,不作跨协议能力变化比较GDPval-AAelo739Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.1GDPval-AA normalizedfraction12.0%Artificial Analysis ↗第三方评测有效
Artificial Analysis intelligence component evaluationsv4.1GPQA Diamondfraction59.4%Artificial Analysis第三方评测有效
Artificial Analysis intelligence component evaluationsv4.1Humanity's Last Examfraction3.6%Artificial Analysis第三方评测有效
Artificial Analysis model aggregatevpublic-model-v1输入价格 / 1M Tokenusd_per_1m_tokensUS$0Model Coordinate repository snapshots第三方评测来源报告 0 / 未验证
Artificial Analysis capability indicesvpublic-model-v1Intelligence Indexindex19.2Artificial Analysis第三方评测有效
Artificial Analysis intelligence component evaluationsv4.1AA-LCRfraction30.0%Artificial Analysis第三方评测有效
Artificial Analysis intelligence component evaluationsv4.1AA-Omniscienceindex-46.2Artificial Analysis第三方评测有效
Artificial Analysis model aggregatevpublic-model-v1输出价格 / 1M Tokenusd_per_1m_tokensUS$0Model Coordinate repository snapshots第三方评测来源报告 0 / 未验证
Artificial Analysis model aggregatevpublic-model-v1生成速度tokens_per_second70.1 Token/sModel Coordinate repository snapshots第三方评测有效
Artificial Analysis intelligence component evaluationsv4.1SciCodefraction33.1%Artificial Analysis第三方评测有效
Artificial Analysis intelligence component evaluationsv4.1Tau Bankingfraction10.3%Artificial Analysis第三方评测有效
Artificial Analysis intelligence component evaluationsv4.1Terminal-Bench v2.1fraction30.3%Artificial Analysis第三方评测有效
Artificial Analysis model aggregatevpublic-model-v1首个回答 Token 等待seconds1.4 秒Model Coordinate repository snapshots第三方评测有效