OpenAI launches GPT-6 Sol and Luna, cutting API prices roughly in half to challenge DeepSeek

APPSO · wechat · 2026-09-23

OpenAI released GPT-6 Sol and GPT-6 Luna with steep API price cuts: Sol drops to $2/$10 per million input/output tokens, Luna to $0.1/$0.5 — about half of the previous generation and, on regular requests, cheaper than DeepSeek V4.1 Flash, though DeepSeek's off-peak cache-hit price ($0.02 per million input tokens) remains unmatched. OpenAI attributes the cuts to inference infrastructure and caching gains, with Sam Altman stressing cost-per-task as the key metric.

Official benchmarks: Sol scored 33.2% on AutomationBench (vs Claude Opus 5's 26.9% at 9% of the cost), 56.4% on Agents' Last Exam, and 68.8% on DeepSWE v1.1 — close to Claude Fable 5's 69.9% at 80% lower cost, though still below DeepSeek V4.1 Flash's 74.2%. Luna matches GPT-5.6 Sol at 1% of the cost. Early hands-on reviews split: one developer reported one-hour tasks finishing in 20 minutes with less than half the tokens, while visual Three.js tests still favored Claude Opus 5.5 on completeness and design taste. Artificial Analysis found intelligence roughly flat vs GPT-5.6, with the real change being cost (per-task cost down nearly half).

OpenAI also upgraded prompt caching (90% discount on cache reads, mid-conversation reasoning adjustments that preserve cache, explicit cache breakpoints). Both models are live in ChatGPT Work and Codex for Plus/Pro/Business tiers and callable via gpt-6-sol and gpt-6-luna, with a one-time usage-quota reset for subscribers. Cheap tokens are no longer any single lab's moat — they're now table stakes at the frontier.

Original post →

OpenAI launches GPT-6 Sol and Luna, cutting API prices roughly in half to challenge DeepSeek

More from Venture