Anthropic announced Opus 5.5 on September 22 in the US. It describes performance near Fable 5.1 on most tasks. Those are release claims; independent task-specific results take longer.
The reported 40% cost reduction concerns typical default-setting workloads versus Opus 5. It is not a blanket 40% reduction in token price or every user’s bill.
McKay welcomes the return of a more enjoyable conversational personality alongside stronger aesthetics. This is an enthusiastic personal assessment, not a paired design study. His usage-limit claim needs plan-specific confirmation.
Editorial paraphrase with a short attributed excerpt · Opus 5.5 · context checked 2026-09-23. Original-source availability may vary.
A C-to-Rust migration puts task completion time in the spotlight.
Cherny reports porting HAProxy with Opus 5.5 and Fable 5.1: nearly all tests passed, with a shorter run and lower cost for Opus in that setup. Published by a vendor employee; we have not audited the harness, patches or failures.
Editorial paraphrase · Opus 5.5 vs Fable 5.1 · context checked 2026-09-23. Original-source availability may vary.
Usage limits are part of the release experience, not a benchmark score.
The release-day post discusses more five-hour capacity and a banked reset. This is a provider statement, not a promise that every account has the same allowance. Check the current rules for your plan.
Editorial paraphrase · Claude subscription product · context checked 2026-09-23. Original-source availability may vary.
Stronger capability does not settle whether safeguards fit a security workflow.
A release-thread reply objects to cybersecurity restrictions. It is a useful counterpoint to launch enthusiasm, but the visible reply supplies no reproducible prompt/response pair; it does not establish a general failure rate.
Editorial paraphrase · Claude safeguards / exact setup undisclosed · context checked 2026-09-23. Original-source availability may vary.
A launch-day account of two models porting HAProxy from C to Rust. The interesting evidence is the whole run and the tests—not just a generated function.
The same brief—a gummy octopus you can grab and stretch in the browser—given to Opus 5.5 and GPT-6 Astra. The creator prefers Opus’s squishier, brighter version and estimates its API cost at $14.01 versus $20.74.
vib3coded · original X post · Claude Opus 5.5 vs GPT-6 Astra · creator-reported; one run each
A playful drill app its author built to try Opus 5.5. After teachers asked about it, the author published the code under MIT—excluding the mascot character, name and logo.
gear_machine (@grmchn4ai) · original X post · Claude Opus 5.5 · creator-reported; settings undisclosed
A deliberately useless website of wobbling, translucent shapes you can poke. The creator says jiggle and transparency effects got much better with Opus 5.5, with fewer tokens.
MatoToushi · original X post · Claude Opus 5.5 · creator-reported; settings undisclosed
These notes were reviewed on September 23. Personal impressions, vendor reports and our selection criteria are labelled separately. We have not run a controlled Opus 5.5 comparison. Unverified release rumors are not model availability.
