Claude Opus 5.5: Fable 5.1 Performance at Opus Price

1. The launch facts

Opus 5.5 arrives three weeks after Fable 5.1 and two months after Opus 5, and it is the first release since Anthropic publicly called for pacing the frontier. It was tested before release by external evaluators including Frontier Design and METR, and on Anthropic’s automated behavioral audit of ~2,000 scenarios it is the strongest-performing model the company has tested to date.

Knowledge cutoff is June 2026. Active status with retirement not sooner than September 22, 2027. Claude Sonnet 5.5 and Claude Haiku 5.5 are promised in the coming weeks, and five-hour usage limits were raised on Pro, Max and Team plans.

2. What actually improved

Three things changed at once: capability per dollar, raw speed, and safety. Anthropic positions it as Fable 5.1-class work — long codebase migrations, multi-repo engineering, computer use — at 40% of the running cost of Opus 5, combining a lower price per token with fewer tokens needed per task.

Early-adopter numbers back the efficiency story. Box reports ~one third fewer tokens than Opus 5 with ~40% shorter response time at equal accuracy. GitHub’s Copilot testing found tasks resolved comparably to Opus 5 with significantly fewer steps and tokens, plus faster recovery from errors in multistep tasks.

The three wins

3. Evidence, with sources

Anthropic’s headline benchmarks are vendor-run, so read them as directional: gains concentrated in coding, agentic and computer-use tasks, math and science reasoning, and long-horizon professional work, matching or exceeding Fable 5.1 and Mythos 5.1 on many evaluations.

The independent signals matter more. Box (early adopter): one third fewer tokens, 40% less latency, no accuracy loss. Copilot (early testing): same task resolution with fewer steps and tokens. Frontier Design and METR ran pre-release external evaluations as part of the pacing-the-frontier commitment.

4. Migration: four breaking changes

The API contract is Fable 5.1’s now. Four changes break code running on Opus 5: thinking cannot be disabled, forced tool use returns an error, thinking blocks are tied to the model and the conversation, and the old computer_20251124 tool is rejected on the Claude API and Google Cloud.

One more change alters response shape without failing any request: text produced between tool calls now arrives inside thinking blocks that are empty at the default display setting. Apps streaming that text as progress updates go quiet between tool calls until they set a display value that returns the text.

model = “claude-opus-5-5”  # was claude-opus-5
# remove: tool_choice = {“any”}  → now returns an error
# remove: thinking = {“disabled”}  → no longer accepted
# if you stream text between tool calls:
# set display = “all” (or a value that returns thinking text)

5. What to do this week

This is a drop-in upgrade for almost every Opus 5 workload, but the thinking and tool-choice changes mean you should touch staging before production. Start with your most expensive agent loop: it is where the 40% lands first.

Verdict

Opus 5.5 is the least debatable kind of release: same-or-better quality on most work, materially cheaper, measurably faster, with the best safety scores Anthropic has published. The breaking API changes are small and well documented.

Migrate staging this week, measure on your workload, and hold one decision open: Sonnet 5.5 and Haiku 5.5 arrive within weeks and may reset the price-performance pick for lighter tiers.