Frontier-Bench v0.1

Agentic terminal-coding benchmark (74 tasks) run via Harbor or mini-SWE-agent; reported at Claude Opus 5 launch.

% solvedHigher is betterCounts toward:Weights: Coding ×0.6The test's websiteOfficial page ↗

The best 4 · 4 models tested

Top 4 · best setting per model · 4 models, 7 results (0 independent)

Measured by independent testersIndependent Reported by the makerVendor-reported

Frontier-Bench v0.1 leaderboard
				Thinking levelSetting
1	Claude Opus 5	Anthropic	44.4%	Extra highadaptive thinking, effort=xhigh	Maker's own figureVendor-reportedAnthropic ↗	mini-SWE-agent (GKE)	24 Jul 2026
2	Claude Fable 5	Anthropic	33.7%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	—	24 Jul 2026
3	Claude Opus 4.8	Anthropic	21.1%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	—	24 Jul 2026
4	Claude Sonnet 5	Anthropic	17.0%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	mini-SWE-agent (GKE)	24 Jul 2026