Frontier Log

Benchmark · InactiveSuperseded by Agents' Last Exam v1.1

Agents' Last Exam v1

Created by UC Berkeley RDI, 3 Jun 2026 · Higher is better · Website

01020304050Jul 2026Oct 202639.5%Gemini 4 Argon29.6%MiniMax M3.1 Flash24.3%Qwen3.8-Flash-Next26.3%Gemini 3.7 Flash27.0%Qwen3.8-Max24.2%Gemini 3.6 Flash13.2%Qwen3.7-Plus11.8%Qwen3.7-Max01020304050Jul 2026Oct 202639.5%Gemini 4 Argon29.6%MiniMax M3.1 Flash24.3%Qwen3.8-Flash-Next27.0%Qwen3.8-Max24.2%Gemini 3.6 Flash

Results

ModelScoreSourceSource check
MiniMax M3.1 Flash19 Oct 2026 · MiniMax29.6%MiniMax M3.1 Flash: Redefining Agent Performance and Cost152 tasks, overall; Claude Code harness; max effortSource image
Gemini 4 Argon30 Sep 2026 · Google39.5%Gemini 4 Argon: our next era of frontier intelligenceALE-Claw harness; 5-hour window; safety filters onSource image
Qwen3.8-Flash-Next26 Aug 2026 · Alibaba24.3%Qwen3.8-Flash-Next: A New Architecture, Towards Ultimate Cost-Efficiencypass@1; harness and subset not statedOur capture
CorroboratedThe same number is printed in 1 other lab document: 1
Gemini 3.7 Flash13 Aug 2026 · Google26.3%Introducing Gemini 3.7 FlashALE-Claw harnessSource image
CorroboratedThe same number is printed in 1 other lab document: 1
Qwen3.8-Max3 Aug 2026 · Alibaba27.0%Qwen3.8-Max: A New Bar for Coding and Coworkharness and subset not stated
Gemini 3.6 Flash21 Jul 2026 · Google24.2%Introducing Gemini 3.7 FlashALE-Claw harness; reported laterSource image
CorroboratedThe same number is printed in 1 other lab document: 1
Qwen3.7-Plus1 Jun 2026 · Alibaba13.2%Qwen3.8-Flash-Next: A New Architecture, Towards Ultimate Cost-Efficiencypass@1; harness and subset not stated; reported laterOur capture
CorroboratedThe same number is printed in 1 other lab document: 1
Qwen3.7-Max20 May 2026 · Alibaba11.8%Qwen3.8-Max: A New Bar for Coding and Coworkharness and subset not stated; reported later

All benchmarks