OpenAI launches GPT-6 Sol and Luna: Sol beats Opus 5 at 9% of the cost
reach_vb · x · 2026-09-23
OpenAI shipped GPT-6 Sol and Luna, bringing Astra's advances to faster, cheaper models with stronger coding/computer use, better factuality, and clearer answers.
Key data:
- On AutomationBench, Sol at xhigh outperforms Claude Opus 5 at max effort at just 9% of the cost per task.
- On OSWorld 2.0 offline, Luna at max beats GPT-5.6 Sol at medium for one-tenth the cost.
- Sol makes about half as many mistakes as GPT-5.6 Sol on internal factuality evals.
- Improved prompt caching: 90% discount on cached input reads; changing reasoning effort or tools no longer breaks the cache.
- Rolling out in ChatGPT Work and Codex (not yet Chat), plus API; Free/Go users get Luna in the desktop app.
API pricing per 1M tokens: Sol $2/$10; Luna $0.10/$0.50.
More from Models
- MKA benchmark blog: how agents are graded on visual professional deliverables — patloeber · 2026-11-12
- New MKA benchmark: Claude Opus 4.5 tops 21 models at 57% on visual knowledge work — patloeber · 2026-11-12
- Google's Physical-Engineering's Last Exam: all 14 frontier models score below 50% — dl_weekly · 2026-11-12
- Commit leak reveals unreleased Gemini 4 variants: Argon, Hydrogen and Helix (unconfirmed) — BlackHC · 2026-11-11
- Rewiring just 89 of 442,368 neurons in Qwen3-8B matches full SFT for a persona — maxsloef · 2026-11-11
- Unsloth ships Qwen3.8-27B Dynamic V3.1 GGUFs with 5-10% accuracy gains — danielhanchen · 2026-11-11