Two cheaper GPT-6 models land in ChatGPT Work for Plus, Pro, Business, Enterprise and Edu. Free and Go get Luna. OpenAI says Astra is still its best model across the board.
A rep's question is never which model is best. It is which model is good enough for a job you run fifty times a week. Until now the honest answer inside ChatGPT Work was to spend the good model on the account brief and accept whatever the cheap one gave you on everything repetitive. Sol and Luna move that line, and OpenAI moved it while saying out loud that Astra is still the better model — which is the more useful half of the announcement. The seller who re-runs their three most-repeated prompts against Luna this week finds out where their own quality floor sits. The seller who switches everything because the price dropped finds out in front of a customer.
A "frontier" model is the most capable tier a lab ships. "Input and output tokens" are how usage is billed through the API: tokens are pieces of words, you pay separately for what you send and what the model writes back, and the figures are quoted per million, which is why they look small. Those API prices are not what shows up on a ChatGPT seat — they matter to whoever builds on the model, not to you in the chat window. "Codex" is OpenAI's coding surface, not a sales one. And "trained with similar methods as GPT-6 Astra" means the same recipe at a smaller size, which normally buys speed and price at some cost in judgment.
The short version: two new cheaper options appeared in the ChatGPT Work model list, and the company that built them says its expensive one is still better.
OpenAI released GPT-6 Sol and GPT-6 Luna on 22 September 2026. In OpenAI's words, the two models "bring frontier intelligence to everyday work with different balances of capability and cost," and were trained "with similar methods as GPT-6 Astra." The ChatGPT release notes carry a same-day entry describing them as model variants in ChatGPT Work and Codex, separate from the Chat models.
Announced: 22 September 2026, on openai.com.
Availability: "GPT-6 Sol and GPT-6 Luna are available in ChatGPT Work and Codex starting today for all Plus, Pro, Business, Enterprise, and Edu users. Free and Go users can access GPT-6 Luna in the desktop app." In the API they are gpt-6-sol and gpt-6-luna. Rollout was staged: "we plan to roll out these models in ChatGPT gradually throughout the day."
Price: API pricing is $2 per million input tokens and $10 per million output tokens for Sol, and $0.10 per million input tokens and $0.50 per million output tokens for Luna — both described as 50% cheaper than their GPT-5.6 predecessors. Whether a ChatGPT Work seat carries any additional cost for them: Not stated.
Regions: Not stated.
What it doesn't do: OpenAI says plainly that this is not an upgrade over its top model — "GPT-6 Astra continues to be our best model across the board," and "While GPT-6 Astra remains the world's best model for computer use, GPT-6 Sol and Luna offer more cost-efficient performance." The announcement names no regions, describes no change to any CRM, email or calling surface, and the comparisons it publishes are coding and automation benchmarks rather than sales work. It does not tell you which of your own jobs the cheaper models are good enough for.
SDR — Your work is the highest-volume and most repetitive on the team, which makes it the best place to test this and the worst place to guess. Take your standing list-research and first-touch prompts, run one day's worth through Luna, and read the output against what you would have sent anyway. If you cannot tell the difference, you have found free speed. If you can, you have learned where your floor is, which is worth more than the model name.
AE — Leave the account brief, the proposal and anything that lands in a customer's inbox on your current model until you have run them side by side on a live deal. OpenAI's own line is that Astra is still its best model across the board. A cheaper tier raises the floor on routine work; it is not a reason to change the model you use before a first meeting with a stakeholder you get one shot at.
Manager — Do not push a team-wide default change off a launch post. Pick two prompts your reps run every day, have two reps run each on both models for a week, and score the one thing that matters: how much they rewrite before it ships. That number tells you whether the cheaper model saves the team anything. The price does not.
RevOps — The API figures are the line worth reading: $2 and $10 per million tokens for Sol, $0.10 and $0.50 for Luna, both half their GPT-5.6 predecessors. Anything you have built on a model endpoint — enrichment, summarisation, routing — is now a cost question you can re-open with real numbers rather than estimates. Confirm what your plan actually exposes in the picker before you promise the team a model they cannot select, and note the rollout was staged across launch day rather than switched on at once.
A cheaper model only pays for itself on work you repeat, so start where the repetition already lives. Promptifi's scheduled prompts are the daily and weekly plays sellers run on a standing basis — run one of them against Luna and against your current model on the same day's pipeline, then compare how much you had to fix in each.
Two minutes, once a week. What changed in AI, and what to run because of it.