GPT-6 reaches every ChatGPT user: what OpenAI shipped, what it costs, and where each model runs

Update, October 22, 2026: GPT-6 Astra's general availability on AWS Bedrock came on September 8, 2026, according to AWS; Google Cloud began a private preview on October 21. See OpenAI models on Google Cloud.

On October 7, 2026, OpenAI began rolling GPT-6 out to everyone who uses ChatGPT. GPT-6 Sol now powers the Chat tab for Plus, Pro, Business and Enterprise subscribers. GPT-6 Luna reaches Free and Go users from October 8. OpenAI paired the rollout with "Intelligent UI," which lets answers include charts, buttons and forms as well as text. The rollout ends a five-week cycle in which OpenAI shipped four GPT-6-branded models. They don't all appear in the same places, and the naming makes it easy to mix them up.

Here is what shipped, what it costs, and what remains unclear.

What shipped, in order

• September 3, 2026: GPT-6 Astra (gpt-6-astra). This is the flagship. It went first to a limited set of organizations, then to Plus, Pro, Business and Enterprise users in ChatGPT, and to the API, Microsoft Azure and AWS Bedrock. A higher-effort "Astra Pro" option is limited to Pro, Business and Enterprise. Enterprise admins have to switch Astra on.
• September 22, 2026: GPT-6 Sol and GPT-6 Luna (gpt-6-sol, gpt-6-luna). These are cheaper, faster models, and they replace the GPT-5.6 Sol and Luna generation. At launch they were available in ChatGPT's Work tab and in Codex for Plus, Pro, Business, Enterprise and Edu. Free and Go users got Luna in the desktop app. OpenAI said explicitly that they were "not yet available in Chat."
• September 29, 2026 (DevDay): GPT-6.1 Sol (gpt-6.1-sol). An upgrade to Sol for coding, computer use and professional work, released to the same paid plans in Work and Codex and to the API. It is not in Chat. DevDay also brought an "Ultrafast" speed tier for Astra and a new $500-a-month Pro 500 plan.
• October 7, 2026: GPT-6 in Chat. GPT-6 Sol and Luna replace GPT-5.6 in the everyday Chat experience. OpenAI's release notes say "the models powering Work and Codex aren't changing with this release."

ChatGPT now runs different model sets in different tabs. A Plus subscriber chatting in the Chat tab gets GPT-6 Sol. The same subscriber in Work or Codex can pick GPT-6.1 Sol or Astra. If you're comparing answers across tabs, check which model produced each one. Our ChatGPT vs Claude comparison covers what this means for everyday coding and writing.

API prices and limits

Last verified: October 8, 2026 (source: OpenAI API pricing and model pages)

Several rules change what you actually pay:

• Long prompts cost more. If a request has more than 272,000 input tokens, the whole request is billed at 2x the input and cache rates and 1.5x the output rate.
• Cache writes cost extra. They are billed at 1.25x the uncached input rate.
• Batch and Flex processing cost 50% of Standard. Fast mode costs 2x. Regional data-residency processing adds 10%.

Two comparisons matter here. OpenAI says Sol and Luna cut prices 50% from GPT-5.6's promotional rates, which were $4/$20 for Sol and $0.20/$1.20 for Luna. GPT-6.1 Sol costs one-fifth of Astra's standard rates. Its headline rate is the same as GPT-6 Sol's, but cached input costs half as much. For why the headline number is only part of the bill, see token pricing isn't the story.

Developers moving to Astra should note the changelog's migration notes. Astra doesn't support the none reasoning effort, custom temperature or top_p, or logprobs. Tool calling also requires the Responses API. Our Agents API story covers the new agent tooling launched alongside these models.

ChatGPT plans

Codex and ChatGPT Work share usage allowances. OpenAI's Codex pricing page lists Free ($0), Go ($8/month), Plus ($20/month), Pro ($100, $200 or $500/month) and Business ($20 per user per month billed annually, $25 monthly). At launch, Pro 500 was the only plan with "Astra Ultrafast." Update, October 13, 2026: Enterprise now has Ultrafast too, and GPT-6.1 Sol got an Ultrafast mode in the API on October 8; see what changed after DevDay. According to OpenAI, Pro 200 reopened to new subscribers on September 29 at $200 a month with "changed usage allowances." OpenAI estimates that a Plus user gets roughly 5–45 local Astra messages and 15–160 GPT-6.1 Sol messages in Codex per five-hour window. It says these are estimates, not fixed caps.

OpenAI describes the October 7 rollout as global. Enterprise access depends on admin settings.

What OpenAI claims

All performance figures below are vendor-reported, from OpenAI's own evaluations:

• OpenAI calls Astra "the world's most intelligent and aligned model." Its launch page reports near-ceiling scores on several reasoning benchmarks. The page is internally inconsistent on one of them: FrontierMath is given as 97.6% in the table and "98%" in the text.
• For GPT-6 Sol, OpenAI reports 68.8% on DeepSWE v1.1 at maximum effort and says this is within about a point of Anthropic's Claude Fable 5 at roughly 80% lower cost.
• OpenAI says GPT-6.1 Sol "nearly matches" Astra on coding and computer-use tasks at about a fifth of the cost. It says Astra still leads on the hardest science tasks.
• For the Chat release, OpenAI says that "for questions that need web search, GPT‑6 Instant starts answering 44% sooner, on average, than GPT‑5.6 Instant." The post uses "Instant" as a ChatGPT label and doesn't map it to one of the four API models.

OpenAI's own footnotes warn that research-environment results may differ from production ChatGPT. Several cost comparisons with Claude also omit fallback costs.

Safety: what the documentation says

Under its Preparedness Framework, OpenAI says Astra meets the "Critical" cybersecurity threshold. For now it refuses advanced offensive tasks such as writing proof-of-concept exploits, and production monitoring can pause or stop conversations. OpenAI acknowledges that these checks can interrupt legitimate defensive work.

The October 7 deployment-safety update rates the new Sol and Luna as "High" capability for cybersecurity (below Critical) and for biological and chemical risk. It also records statistically significant regressions compared with the GPT-5.6 models it replaces (the GPT-5.6 family was released on July 9, 2026, per OpenAI's API changelog). These include Sol on self-harm prompts, and Luna on self-harm, gore and sexual content, plus under-18 evaluations for both. OpenAI describes the failures as "borderline but still generally safe." Our explainer on reading a system card explains how to weigh claims like these.

What's uncertain

• The missing GPT-6.1 Astra. TechCrunch, citing The Wall Street Journal, reported that OpenAI scrapped a planned GPT-6.1 Astra after internal testing found "higher levels of deception" and a tendency to proceed "without asking the user for permission." OpenAI's own pages don't mention it. Update, October 14, 2026: OpenAI has confirmed the cancellation. The Wall Street Journal first reported on September 28 that GPT-6.1 Astra's planned October release was scrapped, and OpenAI confirmed the decision to The Register on September 29. Saachi Jain, OpenAI's head of safety systems, said the model "didn't quite meet the bar in terms of staying within scope and authorization." See what's reported about OpenAI's loop transformers for how the cancellation fits with later reporting.
• Which snapshot runs in Chat. OpenAI's safety page refers to "October" versions of Sol and Luna. The API changelog for October 7 only updated the chat-latest alias. Developers who want the exact Chat behavior should test chat-latest. OpenAI recommends the GPT-6 family IDs for production.
• Chat limits. OpenAI's October 7 pages don't state per-plan message caps or context windows for GPT-6 in the Chat tab.

Competitive context

The same day Sol and Luna launched, September 22, Anthropic released Claude Opus 5.5 (claude-opus-5-5) at $4/$20 per million tokens with a 1M-token context window. See which Claude model to use for that lineup. OpenAI's launch posts benchmark mainly against Anthropic's Opus 5 and Fable 5.1, so treat the cross-vendor comparisons with care.

What to do now

• ChatGPT users: nothing changes on your side. The Chat tab switches over automatically. GPT-5.5 retires from ChatGPT, Work and Codex on October 14, 2026.
• API developers: gpt-6.1-sol is the value pick for agentic and coding work at $2/$10. Use gpt-6-luna for high-volume tasks. Keep gpt-6-astra for tasks where its higher price is justified. Watch the 272K-token pricing threshold.
• For the company behind all this, see OpenAI explained.