Frontier Log

Benchmark · Active

GraphWalks

Created by OpenAI, 12 Apr 2025 · Higher is better · Website

020406080100Apr 2025Jul 2025Oct 2025Jan 2026Apr 202697.7%Mythos Preview75.1%Opus 4.732.4%GPT-5.472.0%Opus 4.650.2%Sonnet 4.525.0%GPT-4.1020406080100Apr 2025Jul 2025Oct 2025Jan 2026Apr 202697.7%Mythos Preview75.1%Opus 4.732.4%GPT-5.472.0%Opus 4.650.2%Sonnet 4.5

Settings and run setups

DomainTask setProviders
BFSUp to 128KOpenAIGoogle
BFS256K + 1M problemsOpenAIAnthropicGoogle
BFS256K problems onlyOpenAIAnthropic
BFS1M problems onlyOpenAIAnthropic
ParentsUp to 128KOpenAI
Parents256K + 1M problemsOpenAIAnthropic
Parents256K problems onlyOpenAIAnthropic
Parents1M problems onlyOpenAIAnthropic

Download image

Results

ModelScoreSource
Claude Mythos Preview7 Apr 202697.7%Claude Mythos Preview System CardChecked A person checked this number against its source on 10 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.5 trials
Claude Opus 4.716 Apr 202675.1%Introducing Claude Opus 4.7Corroborated The same number is printed in 1 other lab document: 15 trials, per the system card · run past the API's 1M-token limitSource image
Claude Opus 4.65 Feb 202672.0%Claude Opus 4.6Corroborated The same number is printed in 1 other lab document: 1max effort, per the system card · 5 trials · run past the API's 1M-token limitSource image
Claude Opus 4.65 Feb 202671.1%Claude Opus 4.6 System CardCorroborated The same number is printed in 1 other lab document: 164K extended thinking · 5 trials · run past the API's 1M-token limit · another report, not charted
Claude Sonnet 4.529 Sep 202550.2%Claude Opus 4.6Corroborated The same number is printed in 1 other lab document: 164K extended thinking, per the system card · reported laterSource image
GPT-5.45 Mar 202632.4%Introducing GPT-5.4Checked A person checked this number against its source on 11 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.Our capture
GPT-4.114 Apr 202525.0%Introducing GPT-4.1 in the APIChecked A person checked this number against its source on 11 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.Run on the set as first published, before the creator reworded the prompts and corrected 24 parents answers on Feb 27, 2026Our capture
GPT-4.1 mini14 Apr 202511.0%Introducing GPT-4.1 in the APIChecked A person checked this number against its source on 11 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.Run on the set as first published, before the creator reworded the prompts and corrected 24 parents answers on Feb 27, 2026Our capture
GPT-4.1 nano14 Apr 20255.6%Introducing GPT-4.1 in the APIChecked A person checked this number against its source on 11 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.Run on the set as first published, before the creator reworded the prompts and corrected 24 parents answers on Feb 27, 2026Our capture

All benchmarks