OpenAI's GPT-6.2 Astra (released November 17, 2026) and Anthropic's Claude Fable 5.2 (released November 9) have the same list price: $8 input / $40 output per 1M tokens. One tier down, GPT-6.2 Sol and Claude Sonnet 5.5 are both $2 / $10.
Short answer: the bills can still differ by up to about 2x. The gap comes from three places: the long-context surcharge, cache pricing, and discount tiers. Sources: OpenAI pricing and Claude pricing, both retrieved November 18, 2026.
Side-by-side pricing (Standard, per 1M tokens)
Item | GPT-6.2 Astra | Claude Fable 5.2 | GPT-6.2 Sol | Claude Sonnet 5.5 |
|---|---|---|---|---|
Input | $8 | $8 | $2 | $2 |
Output | $40 | $40 | $10 | $10 |
Cache read | $0.40 | $0.20 | $0.10 | $0.10 |
Cache write (short) | $10 (30 min) | $10 (5 min) | $2.50 (30 min) | $2.50 (5 min) |
Cache write (long) | $20 (12 h) | $16 (1 h) | $5 (12 h) | $4 (1 h) |
Long-context surcharge | Over 272K: 2x input, 1.5x output | None up to 1M | Over 272K: 2x input, 1.5x output | None up to 1M |
Where the bills split
1. Very long prompts (whole codebases, long contracts)
For one request with 500K input and 20K output tokens: GPT-6.2 Astra = 0.5 × $16 + 0.02 × $60 = $9.20; Claude Fable 5.2 = 0.5 × $8 + 0.02 × $40 = $4.80. That is about 1.9x. Under 272K input tokens, the gap disappears.
2. Re-reading the same documents
Reading a 100K-token knowledge base from cache 10,000 times a month costs $400 on GPT-6.2 Astra vs. $200 on Fable 5.2. But OpenAI's short cache lasts 30 minutes versus Anthropic's 5 minutes, so if requests arrive 5–30 minutes apart, OpenAI may need fewer cache rewrites. Sol and Sonnet 5.5 have the same $0.10 cache read.
3. Work that can wait
Batch is 50% off on both. OpenAI also added a background completion window in the Responses API (beta): 30% off for 2 hours, 40% off for 8 hours. Anthropic's pricing page has no equivalent. Fast mode is $16 / $80 on both flagships; Anthropic's Fast mode cannot be combined with Batch, and OpenAI has an extra Ultrafast tier.
How to choose
- Prompts over 272K tokens? Claude is cheaper.
- Re-reading the same long context often? Fable 5.2's cache read is half price, but compare cache lifetimes.
- Lots of non-urgent jobs? OpenAI's background window saves 30–40%.
- Otherwise, run 20–30 of your real tasks through both and grade the output. With equal list prices, quality and your existing cloud contracts decide.
More on the new OpenAI models: GPT-6.2 Astra and Sol explained.
FAQ
Do GPT-6.2 Astra and Claude Fable 5.2 cost the same?
Their Standard input and output prices are identical at $8 and $40 per 1M tokens. Cache reads differ ($0.40 vs. $0.20), and only GPT-6.2 has a surcharge above 272K input tokens.
Which is cheaper for long documents?
Claude Fable 5.2, when a single request exceeds 272K input tokens. GPT-6.2 then bills 2x input and 1.5x output for the whole request, while Claude keeps the same rate up to 1M tokens.
How do GPT-6.2 Sol and Claude Sonnet 5.5 differ?
Same $2 input, $10 output and $0.10 cache read. The differences are GPT-6.2 Sol's long-context surcharge and cache lifetimes (30 min / 12 h vs. 5 min / 1 h).