OpenAI launched two new models, GPT-6 Sol and GPT-6 Luna, on 22 September, cutting API prices roughly in half against their GPT-5.6 predecessors and pitching them as leaner descendants of the GPT-6 Astra flagship it shipped a few weeks earlier.
The timing was coincidental rather than planned: Anthropic released Claude Opus 5.5 within roughly an hour of OpenAI’s announcement, on a day that followed Grok 4.7 and MiMo v2.6 Flash/Pro the day before. Four frontier labs shipped new models inside 48 hours.
OpenAI says both models make fewer mistakes than their GPT-5.6 counterparts, though it hasn’t published error-rate figures to back that up.
What GPT-6 Sol and Luna change
GPT-6 Luna is OpenAI’s cheapest tier, and the GPT-5.6 version was already popular with developers building applications against the API because it paired strong output with a low running cost. The new model drops input pricing from $0.20 to $0.10 per million tokens, cached input from $0.02 to $0.01 per million, and output from $1.20 to $0.50 per million: a steeper cut on output than on input.
GPT-6 Sol, the mid-tier model, gets a cleaner halving straight down the line: input falls from $4 to $2 per million tokens, cached input from $0.40 to $0.20, and output from $20 to $10. OpenAI describes both models as cut from the same cloth as GPT-6 Astra, meaning they carry over Astra’s underlying capability rather than being scaled-down training runs of their own. AndroidAuthority reports that GPT-6 Sol outperforms the previous-generation Claude Opus 5 while costing just 9% as much to run, and that both new models are live today in ChatGPT Work and in Codex for ChatGPT subscribers.
How Sol’s price compares with Opus 5.5
By our arithmetic, working from the published API rates, GPT-6 Sol prices at exactly half of Claude Opus 5.5 on both ends of the token count: $2 against $4 per million input tokens, $10 against $20 per million output tokens. Opus 5.5 and Sol launched roughly an hour apart, so that precise 2x gap looks less like coincidence and more like OpenAI pricing Sol directly against Anthropic’s still-fresh listing rather than only against its own GPT-5.6 generation.
The wider list shows how far apart the tiers sit once cached-input and output rates are lined up: GPT-6 Luna asks $0.10 and $0.50 per million input and output tokens, GPT-5.6 Terra (OpenAI’s older non-Astra flagship) asks $2 and $12, Grok 4.7 asks $2 and $6, Claude Fable 5.1 asks $10 and $50, and GPT-6 Astra asks $10 per million input tokens on its own. Sol and Luna are priced to undercut everything below Astra, not to compete with it.
What to watch next
Four models from four labs landing inside two days points at accelerating price competition rather than a one-off release cycle. Whether Anthropic answers with a further Opus cut, or OpenAI extends the same treatment to Astra itself, will say more about where this settles than either launch does on its own.
