The news
Anthropic released Claude Opus 5.5 on September 22. The company described the model as the strongest-performing one it has tested to date and the first entry in a new 5.5 model family. Opus 5.5 reaches the level of Claude Fable 5.1 on most work yet costs 40 percent less to run than Opus 5.
Context
Prior Opus models set the price and capability baseline for Anthropic's high-end line. Opus 5.5 narrows the gap between that line and the Fable series by delivering comparable results at lower operating cost. The release also carries safety measures previously reserved for Fable 5.1 because the model performs at a level comparable to Mythos 5.1 on biology and cybersecurity benchmarks.
Anthropic positioned the launch as a direct response to customer feedback on cost and speed for sustained workloads. Teams running repeated agentic sessions or computer-use tasks had previously faced a clear split: pay the higher Opus rate for peak capability or accept Fable's lower ceiling. Opus 5.5 collapses much of that split by matching Fable 5.1 across the majority of evaluated tasks while retaining the faster output speed and reduced per-token expense.
Details
Anthropic reported that Opus 5.5 produces output more than 30 percent faster than Opus 5. On internal benchmarks the model exceeded Fable 5.1 scores in agentic coding, knowledge work, computer use, visual chart recognition, and multidisciplinary reasoning. The company added that the model communicates more naturally than earlier versions and surfaces important information with greater clarity. Safeguards aligned with those on Fable 5.1 accompany the launch because of the biology and cybersecurity parity with Mythos 5.1.
The performance numbers come from Anthropic's own test suites rather than third-party leaderboards. In agentic coding the model completed multi-step repository edits with fewer follow-up corrections than Opus 5 required. Knowledge-work evaluations showed stronger synthesis across long documents, while computer-use trials recorded higher success rates on desktop navigation sequences. Visual chart recognition improved on both accuracy and the ability to explain trends without hallucinated data points. Multidisciplinary reasoning tests combined elements from science, law, and engineering prompts, where Opus 5.5 produced more coherent cross-domain answers.
Safety adjustments mirror the controls already active on Fable 5.1. These include stricter refusal patterns on queries that touch high-risk biology topics or offensive cybersecurity techniques. Anthropic stated the decision followed directly from the model's alignment with Mythos 5.1 scores in those domains. No new external red-team results were published alongside the announcement.
Reactions / counterpoints
Independent benchmark data remains limited. Public leaderboards have not yet incorporated Opus 5.5, so developers must rely on the internal comparisons Anthropic released. Some early users noted that the 40 percent price cut applies only to the new model and does not retroactively change pricing for existing Opus 5 deployments.
Why it matters
Teams that previously chose between Fable's lower price and Opus's higher ceiling now face a narrower trade-off. The 40 percent cost reduction and speed gain make sustained agentic workflows and large-scale computer-use tasks more practical inside existing budgets. At the same time the safety controls imported from Fable 5.1 signal that Anthropic treats the new performance tier as one that requires the stricter guardrails previously applied only to its flagship consumer model. Developers evaluating model swaps will weigh these measured gains against the still-limited public benchmark data released so far.
The shift also changes procurement math for organizations that run inference at volume. A 40 percent reduction in per-token cost, paired with a 30 percent increase in output speed, compounds across thousands of daily calls. Projects that once hit budget ceilings after a few weeks of agentic experimentation can now extend those runs without immediate approval for extra spend. Computer-use features in particular benefit because each desktop action sequence consumes many tokens; the combined price and latency improvements reduce the friction that previously made long-horizon automation expensive to test.
For safety and compliance teams the message is equally direct. The decision to ship Fable-level safeguards on an Opus-class model indicates Anthropic now views any system that reaches Mythos 5.1 biology and cybersecurity scores as carrying equivalent risk. Organizations that had delayed adoption of high-end models because of policy reviews may find the new release easier to clear, since the guardrails already match those accepted for Fable 5.1. At the same time, the absence of fresh external red-team numbers means internal security reviews will still need to run their own tests before granting broad access.
The release also sets expectations for the rest of the 5.5 family. If Opus 5.5 can close most of the gap to Fable 5.1 at lower cost, later models in the same line may follow the same pattern of incremental gains delivered at reduced operating expense. That trajectory would pressure competitors to match both the capability and the pricing discipline rather than relying on flagship premiums alone.
---
Sources:
No comments yet