GPT-6 Luna scores 59.3% on ARC-AGI-2, matching GPT-5.6 at ~62% lower cost
scottleibrand · x · 2026-09-23
ARC Prize published verified ARC-AGI results for OpenAI's GPT-6 Luna:
- ARC-AGI-3: 0.19% ($241, standard harness) / 0.59% ($237, provider adapter harness)
- ARC-AGI-2: 59.3% at $0.062/task
- ARC-AGI-1: 86.7% at $0.018/task
Overall, GPT-6 Luna matches GPT-5.6 Luna's scores at roughly 62% lower cost.
More from Models
- Zvi dissects Anthropic's latest system card, disputing the autonomy threshold call — Don't Worry About the Vase (Zvi) · 2026-11-12
- antirez backs low-bit DeepSeek 4.1 Flash over Qwen 27B dense on Apple M5 Macs — antirez · 2026-11-12
- Gemini re-renders images as text when vision is removed, dropping only 6% WSR — zacharynado · 2026-11-12
- Bare-harness coding benchmark: Fable 5.2 best overall, Gemini 4 Carbon best value — Semantic_meaning · 2026-11-12
- Google's Physical-Engineering's Last Exam: all 14 frontier models score below 50% — dl_weekly · 2026-11-12
- Reddit's 2026 LLM hall of fame: Opus 5.5 still unbeaten after 9 rival releases — G0dZylla · 2026-11-12