What shipped, what it cost on the same four jobs, the effort default Anthropic changed on you, and the four API changes that break Opus 5 code
The companion to "BREAKING: Claude Opus 5.5 Does Fable's Job for a Third of the Price". Anthropic shipped Claude Opus 5.5 on September 22, 2026 at $4 / $20 per million tokens (Opus 5 was $5 / $25; cache reads down 60% to $0.20), raised five-hour usage limits on Pro, Max, Team and seat-based Enterprise, gave subscribers a rate-limit reset they can save, and says it performs at Claude Fable 5.1 level on most work. PAGE 1 — what shipped, the three headline facts, and five things to do today. PAGE 2 — the price table against Opus 5, the limits in Anthropic's own words, how to use the saved reset (Settings → Usage → 'Reset for free' — irreversible, may expire, lost on downgrade), and every place you can run it with the model id. PAGE 3 — our test: the same four real jobs (a bug fix, an off-by-one trap, a customer email, a messy sales CSV), each with an automatic pass/fail check, on Opus 5.5 at medium and high, Opus 5 and Fable 5.1, two runs each. All 32 runs passed; the list-price cost of all four jobs was $0.50 on Opus 5.5 against $1.00 on Opus 5 and $1.43 on Fable 5.1, and Opus 5.5 wrote 98.8 tokens a second against 70.8 and 62.3. The two launch claims that did not reproduce are on the page too. PAGE 4 — the effort dial: Opus 5.5 defaults to medium where Opus 5 defaulted to high, the five levels and when to use each, and how to set it in Claude Code and on the API. PAGE 5 — the four breaking changes for API builders, each with Anthropic's exact error message and before/after code. PAGE 6 — the routing rule: Opus 5.5 at medium every day, high when it stalls, Fable 5.1 for long unsupervised runs at 2.5× the price. Launch-day facts quoted from Anthropic's pages on September 22, 2026; lab numbers are ours, small n, list price. Independent guide from Hyperautomation Labs — not affiliated with Anthropic.
Free. No spam. Unsubscribe anytime.