TL;DR
OpenAI launched GPT-6 Sol and GPT-6 Luna on September 22, 2026, cutting API prices 50% below GPT-5.6 promotional rates and directly challenging both Anthropic's new Claude Opus 5.5 and the rising threat of cheap open-weight models.
What happened
- GPT-6 Sol and GPT-6 Luna join GPT-6 Astra as the second and third members of OpenAI's GPT-6 family, available via API as gpt-6-sol and gpt-6-luna.
- Sol pricing: $2 input / $10 output per million tokens, down from $4 / $20 for GPT-5.6 Sol promotional rates.
- Luna pricing: $0.10 input / $0.50 output per million tokens, making it OpenAI's cheapest capable model to date.
- Both models carry a 1.05 million token context window and support up to 128,000 output tokens, with adjustable reasoning levels from none to max.
- On DeepSWE 1.1, Sol at max reasoning scores 68.8% and Luna scores 66.6%, with Sol also hitting 33.2% on AutomationBench at xhigh reasoning.
Why it matters
- 50% cost reduction passed directly to developers makes high-volume agentic and coding workloads materially cheaper overnight, lowering the economic barrier for enterprise AI deployment.
- Luna at $0.10 per million input tokens undercuts the commodity tier where open-weight models have been winning on price, a deliberate move to keep developers inside OpenAI's ecosystem.
- The same week, Anthropic's Claude Opus 5.5 launched at roughly 40% lower cost than Opus 5, confirming a market-wide race to compress inference costs rather than compete solely on benchmark peaks.
- Prompt caching improvements offer up to 90% off cached input costs, compounding savings for agents and long-context workflows that repeatedly send the same system instructions.
- OpenAI's own internal evaluation claims Sol makes half as many factual errors as GPT-5.6 Sol, though no independent benchmark confirms this, a caveat developers must weigh before committing workloads.
What to watch next
- Whether independent benchmarks validate OpenAI's internal accuracy and coding claims for Sol and Luna, particularly on DeepSWE and agentic evals run by third parties.
- How aggressively Google, Meta, and Mistral respond with price cuts or new open-weight releases now that the $0.10 input token threshold has been publicly set as a commercial target.
- The pace of ChatGPT consumer rollout: Luna is live for Free and Go desktop users now, but broader access and whether Sol reaches standard ChatGPT tiers will determine how fast adoption scales beyond API developers.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.