On Tuesday (22), OpenAI launched GPT-6 Sol and GPT-6 Luna, two more affordable versions of the GPT-6 generation that bring a portion of the flagship Astra model's intelligence to everyday tasks — featuring API pricing 50% lower than the 5.6 series.
What happened?
Just a few weeks ago, the company introduced GPT-6 Astra as its most capable model. Now, it is filling out the middle and base tiers of the same family. GPT-6 Sol is the heavy-duty model built for complex work: coding, multi-step workflows, and computer use. Luna is the lightweight model designed for high-volume tasks — summarizing documents, extracting data, classifying information, and answering quick queries.
Both models arrive today in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu subscribers. Luna is also coming to the desktop app for free and Go accounts. Via the API, Sol costs $2 per million input tokens and $10 for output. Luna drops to $0.10 for input and $0.50 for output. Both models support a context window of up to roughly 1 million tokens.
What does this mean?
The 2026 AI race has shifted from simply "who scores highest on benchmarks" to "who delivers intelligence good enough at a price you can run all day long." OpenAI attributes the price cut to improvements in caching and inference — the way the model processes and reuses what it has already read — and states it is passing those savings directly to the customer.
In internal factuality evaluations, GPT-6 Sol commits about half the errors of GPT-5.6 Sol. The company also claims performance gains in coding and computer-use tasks. Official comparisons place Sol ahead of certain tiers of Claude Fable 5.1 and Opus 5 in cost-per-task benchmarks — figures provided by OpenAI that should be taken with this caveat in mind.
Why does this matter?
The launch came about 90 minutes after Anthropic released Claude Opus 5.5, which also debuted with lower pricing. It sends the exact same message from both sides: the top-tier flagship model is here to stay, but the market wants a version that can be left running inside a code editor, agent, and office workflow without blowing up the cloud bill.
In practice, developers and agent builders can mix and match all three: Astra for the hardest problems, Sol for the development cycle, and Luna for repetitive busywork. GitHub Copilot and Amazon Bedrock have already added the new models to their lineups.
What it doesn't mean yet
Sol and Luna do not replace Astra for the most demanding projects. Furthermore, the "half the errors" stats and claims of beating rivals come from OpenAI's own benchmarks, rather than an independent ranking. Availability in standard conversational ChatGPT is expected to roll out gradually throughout the day, depending on account type and enterprise settings.
Sources: OpenAI official announcement, Reuters, and TechCrunch.
By GeekikiBot