OpenAI released GPT-6 Sol and GPT-6 Luna on September 22, 2026, two cheaper models that sit under September’s GPT-6 Astra. TechCrunch notes the timing: Anthropic had released Claude Opus 5.5 “just 90 minutes before OpenAI’s release.”
The headline is price. OpenAI is “reducing API prices for Sol and Luna by 50% compared with their GPT‑5.6 promotional pricing.” Sol now costs $2 per million input tokens and $10 per million output tokens, half of what Anthropic charges for Opus 5.5.
The catch is where you can use them. In ChatGPT, both models run in ChatGPT Work and Codex, and OpenAI’s announcement says plainly: “These models are not yet available in Chat.” If you open ChatGPT and start typing in the normal chat window, you are not talking to GPT-6 Sol.
Who Gets GPT-6 Sol and Luna
OpenAI’s availability line, in full: “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.”
That breaks down like this:
| Plan | GPT-6 Sol | GPT-6 Luna | Where |
|---|---|---|---|
| Plus, Pro, Business, Enterprise, Edu | Yes | Yes | ChatGPT Work and Codex |
| Free and Go | No | Yes | Desktop app only (not Chat) |
| Regular ChatGPT Chat (any plan) | Not yet | Not yet |
ChatGPT Work, which OpenAI launched in July, is “an agent that can take action across your apps and files, stay with a project for hours if needed, and turn a goal into finished work.” Codex “began as a coding agent for developers.” Chat is the familiar back-and-forth window. OpenAI does not say which mode Free and Go users will find Luna in. Chat is ruled out, and in the desktop app “Chat, Work, and Codex are available on every plan, including Free,” so Work or Codex there is the likely route. MacRumors reads the announcement the same way: “GPT–6 Sol and Luna are not yet available in Chat mode.”
The rollout is staggered. OpenAI says it plans “to roll out these models in ChatGPT gradually throughout the day,” and “if you don’t see the new models in ChatGPT Work or Codex, please try again later.”
Developers got both on launch day. In the OpenAI Application Programming Interface (API), they are gpt-6-sol and gpt-6-luna. GitHub Copilot began rolling out both on launch day, with Sol limited to “Copilot Pro+, Max, Business, and Enterprise plans” and Luna also open to Copilot Pro.
What They Cost
| Per million tokens | GPT-6 Sol | GPT-5.6 Sol | GPT-6 Luna | GPT-5.6 Luna |
|---|---|---|---|---|
| Input | $2 | $4 | $0.10 | $0.20 |
| Output | $10 | $20 | $0.50 | $1.20 |
| Cached input | $0.20 | not listed | $0.01 | not listed |
OpenAI’s table labels both changes “50% cheaper.” For Luna’s output tokens the cut is bigger than that: $1.20 down to $0.50 is a 58% drop. The comparison is against what OpenAI calls GPT-5.6’s “promotional pricing.”
Both models accept a 1,050,000-token context window and return up to 128,000 tokens. Batch processing runs at 50% of standard rates for both models. OpenAI also says it improved prompt caching so agents get “discounts of 90% on cached input-token reads.”
Against Claude. Sol’s $2 and $10 is exactly what Anthropic charges for Claude Sonnet 5, and half of Opus 5.5’s $4 and $20. On cache reads the two are level: $0.20 per million tokens on both Sol and Opus 5.5. MacRumors flags why the per-token gap is not the whole story: Anthropic “says Opus 5.5 uses fewer tokens per task than Opus 5, so it’s not a straight comparison.”
What OpenAI Claims
Every figure in this section comes from OpenAI’s launch post. OpenAI ran its own models; for Claude, it says, “evaluations of competitor models were taken from publicly available reports.”
Fewer mistakes. “GPT‑6 Sol makes about half as many mistakes as its predecessor, approaching Astra-level reliability at much lower cost.” That test is built from conversations where users had already flagged an error, and OpenAI warns these “are not representative of typical usage, where factual errors are more rare.” Luna “at higher effort levels” matches the old GPT-5.6 Sol “at about a hundredth its cost.”
Business workflows. On AutomationBench, a test of end-to-end business workflows using 47 tools, OpenAI publishes this table:
| Model (effort) | Score | Cost per task |
|---|---|---|
| GPT-6 Sol (xhigh) | 33.2% | $0.27 |
| GPT-6 Astra (low) | 30.3% | 3.9x GPT-6 Sol |
| Claude Opus 5 (max) | 26.9% | 11.1x GPT-6 Sol |
| Claude Fable 5.1 with Opus 5 fallback (max) | 31.4% | more than 8.9x GPT-6 Sol |
Computer use. On OSWorld 2.0 (offline set), Sol at its xhigh setting scores 60.5% against 60.3% for Claude Opus 5 at medium, “at approximately 80% lower cost per task.”
Coding. On DeepSWE v1.1, Sol at max effort scores 68.8%, “within 1.1 percentage points” of Claude Fable 5’s 69.9%, “at approximately 80% lower cost per task.”
Neither Launch Chart Includes the Other’s Newest Model
Read the Claude names in those comparisons. OpenAI compares Sol against Opus 5, Fable 5.1 and Fable 5, and says “scores for Claude Fable 5 were reported when scores for Claude Fable 5.1 were unavailable.” Opus 5.5, released 90 minutes earlier, is not in any of them.
Anthropic’s chart has the matching gap. It compares Opus 5.5 against GPT-6 Astra and GPT-5.6 Sol, the predecessor to GPT-6 Sol. So on launch day, no chart from either company puts GPT-6 Sol and Claude Opus 5.5 side by side. The figures in both posts mix each company’s own runs with third-party and rival-reported numbers at different effort settings, so splicing one post’s score onto the other’s table is not a controlled comparison.
AutomationBench shows why. Both posts list Claude Opus 5 at 26.9% and Fable 5.1 at 31.4% (with Opus 5 fallback, in OpenAI’s label). But GPT-6 Astra appears at 30.3% at low effort in OpenAI’s table and at 41.4% in Anthropic’s, where it edges Opus 5.5’s 40.0%. Same model, same benchmark, different settings. Anthropic attributes its Opus 5 and GPT results to “Zapier’s public leaderboard,” which is the place to watch for a like-for-like ranking.
What Else Changed
OpenAI says it brought Astra’s communication style to both models: “more clarity, less jargon, fewer odd turns of phrase, fewer low-value details, and slightly shorter answers overall without losing substance.” In its alignment tests, “both Sol and Luna show improvements over their GPT‑5.6 counterparts, including lower rates of misleading claims about their coding work.”
The lineup is also different from July’s. GPT-5.6 was “a new family of three models: Sol, our new flagship model; Terra, a capable lower-cost option; and Luna, our fastest and most cost-efficient model.” The GPT-6 announcement names only Astra, Sol and Luna, and does not mention Terra. OpenAI still calls Astra “our best model across the board.”
What to Watch
The date that matters to most ChatGPT users is the one OpenAI has not given: when Sol and Luna reach regular Chat. Until then, a Plus subscriber gets GPT-6 Sol only by switching to Work or Codex, and a Free user gets GPT-6 Luna only in the desktop app. For developers, the open question is the head-to-head against Opus 5.5 that neither launch chart ran. The site’s AI model rankings page tracks leaderboard positions and prices as independent results arrive.
Sources (10)
- openai.com Introducing GPT-6 Sol and Luna
- developers.openai.com OpenAI API: GPT-6 Sol model page
- developers.openai.com OpenAI API: GPT-6 Luna model page
- techcrunch.com OpenAI launches GPT-6 Sol and Luna
- macrumors.com OpenAI's New GPT-6 Sol and Luna Models
- anthropic.com Introducing Claude Opus 5.5
- platform.claude.com Claude Docs: Claude Opus 5.5 Overview
- openai.com ChatGPT is now a partner for your most ambitious work
- deploymentsafety.openai.com OpenAI: GPT-5.6 System Card
- github.blog GitHub Changelog: GPT-6 Sol and GPT-6 Luna now available
🦋 Discussion on Bluesky
Discuss on Bluesky