OpenAI launches GPT-6 Sol and Luna: API prices permanently halved, near-frontier scores at far lower cost
赛博禅心 · wechat · 2026-09-23
OpenAI released GPT-6Sol and GPT-6Luna with permanent API price cuts: Sol input $4→$2, output $20→$10; Luna input $0.20→$0.10, output $1.20→$0.50. Both use Astra-like training; Sol targets complex coding/professional work, Luna fast everyday tasks, while Astra remains the strongest. Plus/Pro/Business accounts get a one-time banked reset.
Benchmarks: Sol (max) scores 56.4% on AutomationBench, beating Opus5's best at 60% lower cost per task; on FrontierCode1.1 Main Sol matches Fable5.1 (xhigh) for less; on DeepSWE1.1 (113 tasks, 91 repos) Sol hits 68.8%, 1.1pt behind Fable5's best at 80% cheaper, while Luna (max) at 66.6% is 93–96% cheaper than Opus5/Fable5. On OSWorld2.0 offline Sol (xhigh) scores 60.5% vs Opus5's 60.3% at 80% lower cost. Sol cuts factual errors roughly in half; Luna matches GPT-5.6Sol reliability at 1/100 the eval cost.
Caching & availability: Cache reads now 10% of normal input price, writes 1.25×; new controls include a Prompt Caching Dashboard, mid-run reasoning effort adjustment, tool-scoping and explicit cache breakpoints. GitHub reports reprocessed prompt tokens across billions of OpenAI requests dropped over 50%, making Copilot faster. Available now for paid users in ChatGPT Work and Codex; API names gpt-6-sol / gpt-6-luna, 1.05M-token context.
Related event: OpenAI Launches GPT-6 Sol and Luna at Half the Price(48 posts)→
More from Models
- "People are loving 5.5": Matt Shumer congratulates Anthropic on new model reception — mattshumer_ · 2026-09-23
- Developer claims Anthropic noticed community posts about 'Claudelish' speech quirks — evijit · 2026-09-23
- scaling01 taunts haters after joking Anthropic won't ship Opus 5.5 and Sol today — scaling01 · 2026-09-23
- Delip Rao downgrades from $200/mo Google One Ultra to $50 Pro, leaning on local models — deliprao · 2026-09-23
- Why AI progress accelerated: Claude 4.5 kicked off narrow RSI and open-weight catch-up — maksym_andr · 2026-09-23
- Reward Hacking Traced to Data: Models Reason About LM Graders Leaked in Training Sets — dejavucoder · 2026-09-23