OpenAI releases GPT-6 Sol and Luna at roughly half the previous generation’s API price
On September 23 OpenAI released GPT-6 Sol and GPT-6 Luna, bringing Astra’s training approach to faster, cheaper tiers: Sol costs $2 input and $10 output per million tokens, Luna $0.10 and $0.50.
OpenAI released GPT-6 Sol and GPT-6 Luna on September 23. OpenAI says both use a training approach close to GPT-6 Astra’s and target mainstream work where speed and cost matter. Astra remains the most capable and most expensive tier.
The facts
- Pricing: per ifanr’s breakdown, GPT-6 Sol falls from GPT-5.6 Sol’s promotional $4 / $20 per million tokens to $2 / $10, and GPT-6 Luna from $0.20 / $1.20 to $0.10 / $0.50. For comparison, Astra costs $10 / $50.
- Roles: Sol is for professional work that needs solid reasoning, coding and agent skills. Luna is for high-volume calls.
- Vendor benchmarks: on DeepSWE v1.1, Sol scores 68.8% at max reasoning and Luna 66.6%. On AutomationBench, Sol at xhigh scores 33.2%, at what OpenAI says is about 9% of the per-task cost of Claude Opus 5 at max effort.
- Factuality: testing on anonymized real conversations where users had flagged factual errors, OpenAI says Sol makes about half as many errors as GPT-5.6 Sol.
Our take
Luna puts OpenAI inside the price band DeepSeek had to itself. For ordinary uncached requests it undercuts even DeepSeek V4.1 Flash’s off-peak rate, with no peak pricing. DeepSeek’s cache-hit price is still far lower, though, and V4.1 Flash’s 74.2% on DeepSWE beats Sol.
How to choose: for pipelines that fan out many agents and iterate a lot, Sol and Luna’s savings add up fast. For one-shot front-end or creative work, early side-by-side tests favored Claude Opus 5.5’s output.