OpenAI has spent five weeks turning GPT-6 from a single flagship into a full price ladder. The GPT-6 family now runs from $0.10 to $50 per million tokens, and the rollout reached ChatGPT’s free tier on October 7.
Timeline
| Date | Release | What it is |
|---|---|---|
| Sep 3 | GPT-6 Astra | Flagship at $10 / $50; ARC-AGI-3 99.9%, FrontierMath Tier 4 98%; refuses advanced offensive-cyber tasks |
| Sep 22 | GPT-6 Sol and Luna | Sol $2 / $10 and Luna $0.10 / $0.50, both 50% below GPT-5.6’s promotional prices |
| Sep 29 | GPT-6.1 Sol | ”Near-Astra intelligence for a fifth of the price” at $2 / $10, cached input $0.10 |
| Oct 7 | GPT-6 in ChatGPT | Intelligent UI and GPT-6 Instant in the Chat tab; Sol for paid users, Luna for Free and Go |
All API models share a 1.05M-token context window and 128K max output (gpt-6-astra, gpt-6.1-sol, gpt-6-sol, gpt-6-luna).
What stands out
- The price dropped as the model improved. Sol fell from $4 / $20 (GPT-5.6 promotional) to $2 / $10, and then 6.1 Sol raised capability at the same price. OpenAI’s own numbers have 6.1 Sol matching Astra on DeepSWE v1.1 at about a fifth of the cost.
- A new answer format. Intelligent UI lets ChatGPT respond with interactive diagrams, buttons, forms and on-the-fly tools instead of plain text, for the 1.2 billion-plus weekly users.
- Astra and cyber. Astra reached the critical cybersecurity threshold in OpenAI’s Preparedness Framework; the company is gating the most permissive cyber access behind the Daybreak program while refusing tasks like proof-of-concept exploit writing in the default model.
- Direct pressure on Anthropic and Google. GPT-6.1 Sol’s $2 / $10 price matches Claude Sonnet 5.5 and Google’s introductory price for Gemini 4 Argon, while Astra’s $10 / $50 matches Claude Fable 5.1.
Caveats
- Nearly all benchmark claims are OpenAI-reported, often at xhigh or max effort; independent replication is still thin.
- Astra’s safety checks can pause or stop legitimate work, including defensive security.
- Sol exists as both GPT-6 Sol and GPT-6.1 Sol — new projects should target 6.1.
See the GPT-6 model page for specs, pricing and benchmarks, and the 2026 LLM guide for how it compares.