OpenAI prices GPT-6.2 Astra at $8 / $40, the same as Claude Fable 5.2, eight days after Anthropic moved there
OpenAI launched GPT-6.2 Astra on 17 November 2026 at $8 per million input tokens and $40 per million output tokens, according to its API pricing page as of 18 November 2026. That is 20% below GPT-6 Astra, which my snapshot of the same page, taken on 17 November 2026 before the launch, showed at $10 / $50. It is also exactly the price of Anthropic's Claude Fable 5.2, which Anthropic launched on 9 November 2026 at $8 / $40, down from Claude Fable 5.1's $10 / $50. The two companies' flagship models made the same cut, to the same number, eight days apart.
Flagship and mid-tier prices, as of 18 November 2026
Standard rates in US dollars per million tokens, input / output. OpenAI's figures are its short-context rates.
| Model | Price | Source |
|---|---|---|
| OpenAI GPT-6 Astra | $10 / $50 | OpenAI pricing |
| Anthropic Claude Fable 5.1 | $10 / $50 | Anthropic pricing |
| OpenAI GPT-6.2 Astra | $8 / $40 | OpenAI pricing |
| Anthropic Claude Fable 5.2 | $8 / $40 | Anthropic pricing |
| Anthropic Claude Opus 5.5 | $4 / $20 | Anthropic pricing |
| Google Gemini 4 Carbon from 1 December 2026 | $4 / $20 | Gemini API pricing |
| OpenAI GPT-6.2 Sol | $2 / $10 | OpenAI pricing |
| Anthropic Claude Sonnet 5.5 | $2 / $10 | Anthropic pricing |
Among OpenAI's general models, GPT-6.2 Sol costs the same as GPT-6.1 Sol, so the cut is in the Astra line. The specialist rows moved too: GPT-6.2 Astra for Law lists at $8 / $40 against $10 / $50 for GPT-6 Astra for Law, and GPT-6 Rosalind now lists at $10 / $50, down from $15 / $75 in my 17 November 2026 snapshot.
Same list price, different cache bill
For agents that reread a long prompt on every turn, cache rates decide much of the bill. As of 18 November 2026, per both companies' pricing pages, GPT-6.2 Astra charges $0.40 per million cached input tokens, 60% less than GPT-6 Astra's $1.00. Claude Fable 5.2 charges $0.20 for a cache hit, half of OpenAI's new rate.
| Cache rate, per million tokens, as of 18 November 2026 | GPT-6.2 Astra | Claude Fable 5.2 |
|---|---|---|
| Reading cached input | $0.40 | $0.20 |
| Short cache write | $10 (30 minutes) | $10 (5 minutes) |
| Long cache write | $20 (12 hours) | $16 (1 hour) |
A worked example, by my arithmetic and leaving out cache writes: an agent sends 1 million input tokens, 90% of them read from cache. On GPT-6.2 Astra that costs $0.80 for the fresh tokens plus $0.36 for the cached ones, $1.16. On Claude Fable 5.2 it costs $0.80 plus $0.18, $0.98, about 16% less. Output costs the same on both. OpenAI's long cache write costs more than Anthropic's but keeps the prefix for 12 hours, twelve times longer than Anthropic's one-hour write.
For work that can wait, OpenAI's pricing page lists GPT-6.2 Astra on Batch at $4 / $20, and its changelog added two-hour and eight-hour background windows with rates "30% lower than Standard for the two-hour window and 40% lower for the eight-hour window": $5.60 / $28 and $4.80 / $24 by my arithmetic.
Why OpenAI matched: my reading, and the other one
My reading is that OpenAI did not want its flagship to cost 25% more than Anthropic's, and it had eight days of Fable 5.2 at $8 / $40 to look at. The other explanation is that both companies cut 20% for the same reason, a cheaper model to serve, and landed on the same number because they started from the same $10 / $50. OpenAI's feed describes the launch as bringing "efficiency at lower cost". Either way, GPT-6 Astra stays at $10 / $50, so a team gets the lower price only by moving to the new model ID.
For the market, I read shared price points forming, each held by two vendors. $2 / $10 is OpenAI and Anthropic (GPT-6.2 Sol, Claude Sonnet 5.5). $4 / $20 is Anthropic and Google (Claude Opus 5.5, and Gemini 4 Carbon from 1 December 2026, as I wrote on 17 November 2026); OpenAI has no model there. $8 / $40 is OpenAI and Anthropic. With list prices matched, buyers compare cache rates and tokens spent per task instead.
My forecast: by 31 May 2027, Anthropic or OpenAI will list a new flagship, newer than Claude Fable 5.2 or GPT-6.2 Astra and sold as its most capable general model, at a Standard price under $8 per million input tokens and under $40 per million output tokens. Both companies have now shown they will cut the top price at a new version, and once they sit at the same number, the next launch has reason to undercut rather than match. If on that date neither pricing page lists such a model, I was wrong.