FrontierSWE v2

Proximal's 34 ultra-long-horizon engineering/research tasks (~20h per task), run in Proximal's harness at max effort.

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

The best 8 · 8 models tested

Top 8 · best setting per model · 8 models, 8 results (2 independent)

Measured by independent testersIndependent Reported by the makerVendor-reported

FrontierSWE v2 leaderboard
				Thinking levelSetting
1	GPT-6 Astra	OpenAI	65.5%	Maxreasoning effort=max	Independent testIndependentProximal (via Anthropic system card) ↗	Proximal harness	22 Sep 2026
2	Claude Opus 5.5	Anthropic	62.3%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	Proximal harness	22 Sep 2026
3	Claude Sonnet 5.5	Anthropic	61.9%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	Proximal harness	28 Sep 2026
4	Claude Fable 5.1	Anthropic	56.3%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	Proximal harness	22 Sep 2026
5	Gemini 4 Argon	Google (Gemini / DeepMind)	55.0%	Maxhighest thinking settings	Maker's own figureVendor-reportedGoogle ↗	—	30 Sep 2026
6	Claude Opus 5	Anthropic	52.0%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	Proximal harness	1 Sep 2026
7	Claude Fable 5	Anthropic	48.0%	Maxadaptive thinking, effort=max	Maker's own figureVendor-reportedAnthropic ↗	Proximal harness	1 Sep 2026
8	GPT-5.6 Sol	OpenAI	32.2%	Maxreasoning effort=max	Independent testIndependentProximal (via Anthropic system card) ↗	Proximal harness	22 Sep 2026