OpenAI Cut GPT-6 Sol and Luna's Prices in Half. Astra Isn't Part of the Discount.

What happened

OpenAI expanded its GPT-6 family with two new models, Sol and Luna, and cut their prices in half compared to the previous generation’s promotional rates. Tokens are the small chunks of text a model reads and writes; API pricing is quoted per million of them.

GPT-6 Sol now costs $2 per million input tokens and $10 per million output tokens, down from GPT-5.6 Sol’s $4 and $20. GPT-6 Luna costs $0.10 per million input tokens and $0.50 per million output tokens, down from GPT-5.6 Luna’s $0.20 and $1.20. OpenAI says both were trained with methods similar to GPT-6 Astra, the flagship model it introduced on September 5, but built to run cheaper and faster. Astra’s own pricing, $10 input and $50 output per million tokens, is unchanged.

The same day, GitHub added both models to GitHub Copilot’s model picker. GitHub describes Sol as “a balanced model for interactive and agentic coding” and Luna as “a lightweight, cost-efficient model for smaller, faster tasks.” Sol is available to Copilot Pro+, Max, Business, and Enterprise plans. Luna is also available to the base Pro plan. Both are billed at usage-based rates under GitHub’s model pricing, and the rollout is gradual, so an eligible account may not see the new models immediately.

Why it matters

Before this, choosing an OpenAI model in the GPT-6 family mostly meant choosing Astra, a model built for long, hard, multi-step work and priced accordingly. Sol and Luna give you two cheaper tiers instead of one expensive one, which matters because most coding work isn’t the long, hard kind. A routine edit, a small bug fix, or a quick lookup doesn’t need the model that costs five times as much per token.

That’s the real shift: OpenAI is now selling a ladder, not a single model. Astra for work where getting it right the first time matters more than cost. Sol for everyday agentic coding where you still want careful, multistep checking. Luna for fast, small tasks where speed and low cost matter more than depth. If you’re picking a model inside Copilot or a similar tool without thinking about which tier fits the task, you’re likely paying more than you need to for routine work, or under-provisioning a task that actually needed Astra’s persistence.

Who should care

Anyone choosing a model inside GitHub Copilot’s model picker, and anyone whose coding agent or app builder connects to OpenAI’s API, the direct line software uses to send requests to OpenAI’s models instead of going through a chat app, and lets you pick which model handles the request. If your admin manages Copilot for a team, the same model policy setting that controlled Astra’s rollout controls Sol and Luna too: new models are enabled by default unless an administrator has turned that off or blocked a specific model.

What builders should do next

Don’t assume Astra is always worth the extra cost, and don’t assume Luna is always cheap enough. Pick one task you run often, such as fixing a failing test or updating a function to match a new interface. Run it three times: once each on Astra, Sol, and Luna. Compare how many follow-up corrections each one needed before the change was actually right, and compare the token cost each run used based on the pricing above. If Sol gets you a correct result with one or two corrections at a fifth of Astra’s cost, that’s your default for that kind of task. Save Astra for work where an incorrect first attempt is expensive to catch and fix.

Where this fits

OpenAI is following a pattern other model vendors already use: one flagship model for the hardest work, and cheaper siblings trained the same way for everything else. The Copilot rollout puts that choice directly in front of builders who may never have compared per-token pricing before. Worth checking which tier you’re actually using by default, since the answer might be costing you more than the task requires.

End of article