Anthropic released Claude Opus 5 on Friday, July 24, a new version of its long-running Opus line that the company says comes close to the intelligence of its top-tier Fable 5 model at half the price. The launch is Anthropic's fourth major model release in under two months, and it repositions Opus as the sensible default for most paying customers.
The model is available now on Claude.ai, Claude Code, Claude Cowork, and the Claude API under the identifier claude-opus-5. It becomes the default model on the Claude Max subscription and the most capable option on Claude Pro.
The Benchmarks Tell an Unusual Story
New models normally slot in below the flagship. Opus 5 does not.
On Frontier-Bench v0.1, an agentic coding evaluation, Anthropic reports that Opus 5 surpasses every other model, including Fable 5, while more than doubling the score of its predecessor Opus 4.8 at a lower cost per task. On CursorBench at maximum effort, it lands within 0.5 percent of Fable 5's peak score at half the cost. On ARC-AGI 3, a test of novel problem solving, its score is three times that of the next best model.
The pattern repeats on business-oriented tests. Opus 5's pass rate on Zapier's AutomationBench, which measures whether a model can carry a real task from start to finish, runs about 1.5 times the next best model at matching spend, and even its lowest effort setting passes more tasks than any competitor. On the OSWorld 2.0 computer-use benchmark, it beats Fable 5's best result at just over a third of the cost.
Anthropic also highlighted the model's persistence. In one benchmark task, Opus 5 was asked to rebuild a machine part as a 3D CAD model from a drawing it had been given no way to view directly, so it wrote its own computer vision pipeline to pull the geometry out of the raw pixels and finished the reconstruction. No competing model solved that task in five attempts.
Pricing Stays Flat While Capability Jumps
API pricing is unchanged from Opus 4.8: $5 per million input tokens and $25 per million output tokens. A Fast mode runs about 2.5 times quicker at double the base price. Anthropic's announcement frames the model as one built for daily professional use rather than occasional heavyweight jobs, with Fable 5 still recommended for the longest autonomous work.
An adjustable effort setting lets developers trade reasoning depth for speed and token savings on a per-request basis, and much of the claimed cost advantage sits there. Several launch charts show Opus 5 matching or passing Fable 5's scores at a fraction of the spend.
Fewer Guardrail Interruptions, Plus a Fallback System
Safety restrictions were a persistent complaint about Fable 5, particularly among security researchers. Anthropic has loosened them here: the company expects Opus 5's cyber classifiers to intervene around 85 percent less often than Fable 5's. The model can hunt for vulnerabilities in source code, a task more associated with defensive work, but it is blocked from generating exploits or running penetration tests, and it cannot scan compiled binaries for flaws.
A new fallback system softens the restrictions that remain. Inside Anthropic's own apps, flagged requests now fall back to Opus 4.8 by default instead of returning an error, and API customers can enable a beta called automatic fallbacks that reroutes blocked prompts on either Opus 5 or Fable 5 to another model. Biology requests blocked on Fable 5 now route to Opus 5, which Anthropic describes as its most capable generally available model for scientific research. Members of the company's Cyber Verification Program get a version with fewer security restrictions from day one.
One further detail matters for privacy-conscious users: Opus 5 carries no data retention requirement for general access, unlike the 30-day policy applied to Fable 5 and Mythos 5.
Alignment Scores and a Crowded Release Calendar
Anthropic's automated behavioral audit rated Opus 5 as its most aligned model to date, with the lowest recorded rates of deceptive behavior and the least susceptibility to being tricked into misuse. The company says it deliberately avoided training the model on offensive cyber tasks, and while Opus 5 nearly matches the restricted Mythos 5 at finding software vulnerabilities, it trails far behind at turning them into working exploits.
The release caps a compressed shipping schedule. Opus 4.8 arrived on May 28, and June brought three more models: Mythos 5, Fable 5, and Sonnet 5. Only Haiku, the lightweight tier, still awaits a 5-series upgrade. With Opus 5 now undercutting the flagship on price while beating it on several headline benchmarks, the open question is what buyers of the twice-as-expensive Fable 5 are paying for.
Comments
Join the discussion and share your perspective.