Anthropic launches Claude Opus 5.5: 40% cheaper on typical workloads, five-hour usage caps lifted

Start with money. Per Quartz, Opus 5.5 is listed at $4 per million input tokens and $20 per million output tokens, 20% below Opus 5's $5 and $25, with cache reads down to $0.20 per million tokens from Opus 5's $0.50, a 60% reduction. Quartz specifically notes that cache reads make up the majority of costs in agentic and coding workloads, so the cut there often moves the real bill more than the headline rate. Anthropic's own framing is that because completing the same work requires fewer tokens, typical workloads cost about 40% less to run than on Opus 5. This is the part most easily misread: 40% is a per-task number and 20% is a per-token number, and the two are not in conflict; they are two measurements of one fact, with part of the saving coming from price and the rest from using fewer tokens.

The second thread is capability positioning, and there are two sets of numbers to keep apart: published benchmarks and internal tests. Per MacRumors, on Anthropic's published benchmarks Opus 5.5 surpassed Fable 5.1 on agentic coding, knowledge work, computer use, visual chart recognition and multidisciplinary reasoning. Anthropic also says the model communicates more naturally, putting important information at the front of a response so it is easier to understand at a glance, is less likely to use jargon or idiosyncratic phrases, and follows the writing rules it is given. But Quartz reports that Anthropic itself acknowledged that at current capability levels, benchmark margins have become a less reliable guide to real-world differences, and that the gap between the two models is narrower in practice than the scores suggest. That kind of self-discounting on one's own benchmarks is not common in a launch post.

The internal results are more concrete and say more about where this launch puts its weight. Per Quartz, Anthropic ran an internal coding test asking both models to translate widely used server software from C into Rust: Opus 5.5 completed the task in 9.5 hours at 51% lower cost, while Fable 5.1 took 12 hours. Early tester feedback is in the report too: one tester used Opus 5.5 to audit and fix a 200,000-line codebase in under three hours, where the same job took Opus 5 more than 20 hours and consumed 2.5 times as many tokens, and investment firm Walleye Capital said the model cleared their evaluation suite on its lowest setting and, at higher settings, identified an error in their evaluation instructions that no prior model had caught. Taken together, the pitch is not smarter but equally smart and cheaper: the headline is efficiency, not a new ceiling.

The fourth thread is the change subscribers feel directly. Per MacRumors, with Opus 5.5 live, Anthropic raised the five-hour usage limits on Pro, Max, Team and seat-based Enterprise plans, and subscribers also received a rate limit reset usable any time between now and October 22. Quartz adds that Anthropic expects to ship Claude Sonnet 5.5 and Claude Haiku 5.5 within the next few weeks, which makes 5.5 a family rather than a single model. Benzinga reports the five-hour usage caps were removed on those plans. For heavy daily users this may matter more than any benchmark number, since limits decide how long a run you can sustain.

The fifth thread is the safety handling, which matters as much as the launch itself. Per MacRumors, because Opus 5.5 is comparable to Claude Mythos 5.1 in biology and cybersecurity capability, Anthropic is launching it with safeguards similar to those on Fable 5.1: most cybersecurity tasks will be rerouted to an older model, while vetted researchers can apply to Anthropic's Cyber Verification Program and Life Sciences Verification Program for broader access. Quartz reports that before launch, two independent safety research organizations, Frontier Design and METR, put the model through external evaluations; on Anthropic's internal behavioral audit spanning nearly 2,000 simulated scenarios, it outperformed every previous Claude; and in a dedicated containment evaluation, Anthropic says Opus 5.5 showed roughly an 85% lower rate of boundary-bypass attempts compared with Opus 5. These are vendor-reported figures, but at least they come with named metrics you can argue with.

Finally, there is a contrast in the timeline that may be the most memorable part of this launch. Gizmodo reports that less than two weeks after Anthropic chief executive Dario Amodei urged the industry to pace the frontier and slow down before it loses control of the technology, the company released its most powerful model to date. Quartz, citing TechCrunch, notes this is the first release since Amodei publicly called for slowing the pace of capability development so safety practices can keep up. Investing.com reports the launch also comes ahead of an expected IPO, with Anthropic under competitive pressure from OpenAI. Put together, the logic is not complicated: one track is public policy argument, the other is market competition, and the product release calendar is set by the latter.