OpenAI Cuts GPT-6 Sol and Luna Prices in Half

OpenAI cut API prices for two of its GPT-6 models on Tuesday, dropping GPT-6 Sol to $2 per million input tokens and $10 per million output, and GPT-6 Luna to 10 cents and 50 cents, each half of what the company charged for the equivalent GPT-5.6 models. The announcement, posted to OpenAI's own developer forum, frames both as carrying forward training methods OpenAI first used on GPT-6 Astra, its top-tier model released Sept. 10.

OpenAI said Sol "makes about half as many mistakes as its predecessor, reaching Astra-level reliability at much lower cost," and that cached input tokens now carry a 90 percent discount off the standard rate. The company did not publish a context window figure for either model, and named no executive or customer in the announcement itself.

What the benchmarks claim

On OSWorld 2.0, a test of computer-use tasks, OpenAI scored Sol at its highest reasoning setting at 60.5 percent, just ahead of Claude Opus 5 at medium effort's 60.3 percent, at what OpenAI said was roughly 80 percent lower cost. On DeepSWE v1.1, a software-engineering benchmark, Sol reached 68.8 percent against 69.9 percent for Claude Fable 5 at its highest effort setting, again at a large cost gap OpenAI put at about 80 percent. Luna, the cheaper of the two, scored 66.6 percent on the same test, which OpenAI said is comparable to Opus 5 and Fable 5 running at medium effort.

1. Claude Fable 5 (xhigh)69.9 %
2. GPT-6 Sol (max)68.8 %
3. GPT-6 Luna (max)66.6 %

Source: OpenAI, Introducing GPT-6 Sol and Luna, and Techzine, accessed 2026-09-23

Model	Input, per 1M tokens	Output, per 1M tokens
GPT-6 Luna	$0.10	$0.50
GPT-5.6 Luna	$0.20	$1.20
GPT-6 Sol	$2	$10
GPT-5.6 Sol	$4	$20
GPT-6 Astra	$10	$50

Same Tuesday, a rival cut its own price

Sol and Luna shipped hours after Anthropic released Claude Opus 5.5 at $4 and $20 per million tokens, a 20 percent cut from Opus 5. On a straight base-rate comparison, Sol is still cheaper than Opus 5.5 even after Anthropic's move. Anthropic named customers in its own release; GitHub's chief product officer, Mario Rodriguez, said in materials Anthropic published that Opus 5.5 "solved more terminal tasks than Opus 5 in less than half the steps." OpenAI's post for Sol and Luna carries no equivalent customer or staff quote, Techzine reported the same day, in a write-up that also confirmed OpenAI's full price table for both new models and the GPT-5.6 models they replace.

Who gets it and when

GPT-6 Sol and Luna are live today in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu accounts, with Luna also reaching Free and Go users in OpenAI's desktop app. Developers can call the models directly as gpt-6-sol and gpt-6-luna in the API. OpenAI described the rollout to the ChatGPT app and website as gradual, without saying when it completes.

This extends the run of coverage on GPT-6 Astra's own staged rollout and on OpenAI's prior GPT-Live-1 price cut, OpenAI's second and third API price reductions in as many months. The company has not said whether GPT-6 Astra itself will get a similar cut.