Claude Opus 5.5
- ID
- 27381
- Status
- summarized
- Published
- 23 Sep 2026, 12:29 AM
- Fetched
- 24 Sep 2026, 2:03 PM
- Provider
- Hacker News
- Category
- dev-community
- Original URL
- https://www.anthropic.com/claude-opus-5-5
- Source URL
- https://hnrss.org/best
Summary
- Score
- 8.0
- Created
- 24 Sep 2026, 2:03 PM
- Tags
- Audience
- developersvibe_codersai_ml_learnersai_agent_userssaas_founders
What happened
Anthropic introduced Claude Opus 5.5 on September 22, 2026, the first model in its new Claude 5.5 family, claiming performance at the level of Claude Fable 5.1 on most work at 40% lower cost to run than Opus 5. Token pricing is $4 per million input and $20 per million output (20% below Opus 5), cache reads drop to $0.20 per million (60% below Opus 5), and output generates more than 30% faster. Anthropic says external evaluators including Frontier Design and METR tested it before release, that it scored best to date on its internal automated behavioral audit, and it cites a tester completing a 680,000-line code migration in under a day and fixing page load times in 39 of 40 attempts where Opus 5 made smaller changes that altered app behavior.
Why it matters
The cache-read price is the number to act on: Anthropic states cache reads make up the majority of agentic and coding work costs, and $0.20 per million versus Opus 5's price cuts that line item by 60% — so agent loops that re-read large contexts get materially cheaper per run, and anyone building agent products with USD token costs against local pricing should re-run their unit economics rather than assume last quarter's numbers. The migration and 39-of-40 load-time claims are self-reported and not accompanied by published benchmark tables here, so treat them as a hypothesis to A/B on your own repo before switching default models. Note also that biology and cybersecurity use now runs through verification programs — the Life Sciences Verification Program is open to apply today, with Cyber Verification Program access expanding in the coming weeks — so teams in those domains can't just flip a flag.
Discussion angle
Anthropic frames Opus 5.5 as its first release since it called for pacing the frontier, and the strongest-performing model on its own automated behavioral audit — but the external evaluators named (Frontier Design, METR) have no numbers in this post, so ask whether a lab's self-reported alignment suite plus a 60% cache-price cut is enough to move your default model, or whether you'd want independent evals first.