Claude Opus 5.5 Launch Cuts Costs 40% While Matching Fable 5.1

Anthropic makes its flagship cheaper

Anthropic released Claude Opus 5.5 on September 22, 2026. It is the first model in the new Claude 5.5 family, and the company says it performs at the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5.1 The model is available through Anthropic's own platform and on Amazon Web Services, Google Cloud and Microsoft's cloud. Developers call it with the identifier claude-opus-5-5.12

Four weeks later, the release looks less like a one-off upgrade and more like the start of a reorganized Claude lineup. Sonnet 5.5 arrived on September 28 and Haiku 5.5 followed on October 7. That fills out the 5.5 family within about two weeks of the Opus launch.1112 Taken together, the launches show what Anthropic is betting on: frontier-level capability that is cheap enough to be the default, not a premium tier.

What actually shipped

Opus 5.5 has a 1-million-token context window, up to 128,000 output tokens per response, and a reliable knowledge cutoff of June 2026.2 On Anthropic's Batch API, a beta header raises maximum output to 300,000 tokens.2

Pricing is the biggest change. Input costs $4 per million tokens and output costs $20 per million, both 20% below Opus 5's $5 and $25.25 Cache reads, which Anthropic says account for most of the cost in agentic and coding work, fell 60% to $0.20 per million tokens.1 Anthropic attributes the price cut to the model needing less compute to serve. It also claims output is generated more than 30% faster than on Opus 5.13

The 40% headline figure is a blend: lower per-token prices combined with Anthropic's claim that the model finishes tasks using fewer tokens.10 Customer statements published by Anthropic make the same point from the user side. Optiver says Opus 5.5 matched Opus 5's quality in about half the turns, time and output tokens. Box says it used a third of the tokens Opus 5 did.4

Consumers saw changes too. Opus 5.5 is available on the Pro, Max, Team and Enterprise plans, and Anthropic raised the five-hour usage limits on those tiers. Subscribers also received a rate-limit reset they can use any time before October 22.47 The Free plan does not include the model.10

The benchmark story, and who is telling it

Anthropic says Opus 5.5 is "the new leading model."1 The company's comparisons focus on cost per task as much as raw scores. On Terminal-Bench 4.0, Anthropic says Opus 5.5 at default effort beats Opus 5 at maximum effort for about a fifth of the cost, and matches OpenAI's GPT-6 Astra for about 40% of the cost.1 On the GDPval-AA v2.1 knowledge-work test, which covers 44 occupations, it scored 1846 Elo, ahead of both Fable 5.1 and Opus 5.1 Other reporting adds a 67.7% score on Humanity's Last Exam with tools, which it describes as a 10.5-point lead over GPT-6 Astra.9

The headline performance figures come from Anthropic and from customers the company chose to quote. Independent confirmation is thinner, but it points the same way. One overview says Opus 5.5 at maximum effort took first place on Artificial Analysis's Intelligence Index with a score of 58 as of September.10 The same piece notes that the GPT-6 Astra figure in Anthropic's Terminal-Bench table was reported by OpenAI at a different effort setting.10 Readers should weigh the head-to-head comparisons with that caveat in mind.

The coverage mostly agrees, with one small mismatch. Anthropic described an internal test in which Opus 5.5 and Fable 5.1 each rewrote the load balancer HAProxy from C into Rust. Opus 5.5 finished in 9.5 hours against Fable's 12 and cost 51% less.3 One financial outlet described this as a "51% improvement" over Fable's time, which mixes up the cost saving with the time saving.9 The actual gap in time is closer to 20%. The practical point holds either way: the cheaper model beat the more expensive one on both speed and cost.

The safety framing is the real Anthropic angle

Anthropic is presenting Opus 5.5 as a safety milestone as much as a capability one. The company calls it its first release since it "called for pacing the frontier."1 Outside evaluators, including METR and Frontier Design, tested the model before release. Anthropic says it scored best of any model it has tested on its automated behavioral audit, its most thorough alignment test.13

Anthropic is also open about how capable the model is in sensitive areas. It says Opus 5.5 is comparable to Claude Mythos 5.1, its restricted-access model, in biology and cybersecurity, and so it ships with safeguards similar to those on Fable 5.1.1 According to Anthropic, this is the first Opus model to launch with that class of protections, which also covers anti-distillation measures.4 One report says the system card runs to 230 pages and that the model has the strongest cyber capabilities of any model Anthropic has released publicly.9

That makes access a policy question as well as a pricing one. Vetted organizations can apply to a Life Sciences Verification Program to use Opus 5.5 for biology research.1 On October 6, the Cyber Verification Program was split into Defense, Red Team and Specialized access tiers. It now folds in Project Glasswing and gives qualifying security professionals access to Opus 5.5, Sonnet 5.5 and Mythos 5.1, with safeguards specific to each tier.13 Mythos 5.1 was opened to approved program members the same day.19

The resulting structure is notable. The public gets a model whose most dangerous capabilities are filtered, while verified specialists get gated, less-restricted access. Anthropic first used this split with the Mythos and Fable models, and it is now extending it into its mainstream Opus line.1219

One technical change shows how product design and safety policy overlap. On Opus 5.5, adaptive thinking is always on. Requests that try to disable it return an error, and developers control reasoning depth with an effort parameter instead.220 Anthropic's documentation presents this as one of four breaking changes for code built on Opus 5. Forced tool use now returns an error, thinking blocks are tied to the model and conversation, and an older computer-use tool is no longer accepted on some platforms.2 One analysis reads the always-on thinking requirement as an anti-distillation measure.9 Anthropic does tie anti-distillation to Opus 5.5's safeguards in general, but whether this specific change serves that purpose is that outlet's interpretation.4

Why the pricing matters competitively

Coverage differs most on why the price came down. Anthropic's explanation is technical: the model is cheaper to serve.1 Yahoo Finance's analysis instead places the cut in a market under pressure, citing Grok 4.7's $2/$6 pricing and the open-weight competition from Xiaomi's MiMo-V2.6.9 When Haiku 5.5 launched, Yahoo Finance described the AI pricing war as intensifying.18

Both explanations are probably true, and they support each other. The release pattern backs up the competitive reading. Sonnet 5.5 cache reads were cut from $0.20 to $0.10 per million tokens nine days after it launched.13 Haiku 5.5 arrived at $0.10 per million input tokens for prompts under 100,000 tokens.16 Anthropic is also now steering most workloads to Opus 5.5 first, ahead of the more expensive Fable 5.1.10 That is a deliberate move to make its most capable everyday model the default, rather than holding it back for premium buyers.

Customers adopted it quickly. Agent platform CellCog says it moved every tier that ran on Fable 5.1 over to Opus 5.5 on launch day.16 Developer tools added the model within the first week.10

The cadence question

The release pace adds context. Opus 5 launched on July 24, so Opus 5.5 came about two months later, which 9to5Mac notes is in line with Anthropic's recent rhythm.311 Fable 5.1 and Mythos 5.1 came out three weeks earlier, on September 1.11 Anthropic says Opus 5 remains available until at least July 2027, and Opus 5.5 will be supported until at least September 22, 2027.210

All of this has to be read alongside Anthropic's own call to pace the frontier. Shipping four new models in five weeks does not look like slowing down. Anthropic's apparent argument is that pacing means controlling access to dangerous capabilities, not limiting how often it ships. Opus 5.5 is the clearest example so far. It brings Fable-class performance to a mass-market price and puts Mythos-level biology and cyber capability behind a verification process. The open question is whether those gates hold up as the capabilities spread more widely and get cheaper.