The most interesting number in Anthropic’s Claude Opus 5 launch is the one that did not change. The new model, released July 24, keeps its predecessor’s exact API pricing — $5 per million input tokens and $25 per million output tokens, the same as Opus 4.8 — while Anthropic claims it “comes close to the frontier intelligence of Claude Fable 5 at half the price.” In a market where each frontier release usually arrives with a new bill attached, holding the line on price is itself the pitch.
Opus 5 lands roughly two months after Opus 4.8, matching the cadence Anthropic has kept through 2026, as 9to5Mac notes in its coverage of the release. It becomes the new default model on Claude Max and the strongest model available on Claude Pro, and it is live today across every platform where Claude runs.
Where it sits in Anthropic’s increasingly crowded lineup
Anthropic’s positioning is unusually candid about the pecking order. Opus 5 is the new state of the art on coding and knowledge-work evaluations like Frontier-Bench v0.1 and GDPval-AA — but the company concedes it “remains behind Mythos 5 on cybersecurity tasks.” Fable 5, for reference, is the released, safeguarded version of the Mythos model that Anthropic initially held back over concerns about its cybersecurity capabilities, and it remains the company’s most capable model overall.
So the ladder now reads: Haiku for cheap and fast, Sonnet in the middle, Opus 5 as the everyday workhorse, and Fable 5 at the frontier. Opus 5’s job is to make the frontier tier feel optional.
The benchmark claims backing that up are aggressive. On Frontier-Bench v0.1, Anthropic says Opus 5 surpasses every other model and more than doubles Opus 4.8’s performance at a lower cost per task. On CursorBench 3.2 at max effort, it lands within 0.5 percent of Fable 5’s peak score at half the cost per task. On ARC-AGI 3, a novel-problem-solving evaluation, Anthropic reports a score three times as high as the next-best model, and on the OSWorld 2.0 computer-use benchmark it says Opus 5 beats Fable 5’s best result at just over a third of the cost. These are vendor-published numbers on vendor-selected benchmarks — the usual caveats apply — but the consistent theme is performance-per-dollar rather than raw peak capability.
The behavior Anthropic is actually selling
Beyond the leaderboard talk, Anthropic’s launch notes emphasize how the model works rather than what it scores: Opus 5 is described as “much stronger at verifying its work and iterating carefully until it succeeds.” The company’s showcase example is telling. Given a Frontier-Bench task to rebuild a machine part as a 3D FreeCAD model from a drawing it was deliberately given no way to view, Opus 5 wrote its own computer-vision pipeline to extract the geometry from raw pixels — repeatedly succeeding where, Anthropic says, no competing model solved it in five attempts.
Early-access customers echo the framing. Zapier CEO Wade Foster says Opus 5 topped the company’s AutomationBench leaderboard and ran a full churn-prevention workflow end to end, hitting 100 percent where previous models did not pass. Cognition’s Scott Wu and Cursor co-founder Sualeh Asif both describe near-Fable performance at Opus cost inside their coding tools.
Anthropic also reports measurable gains for scientific work: Opus 5 scores 10.2 percentage points higher than Opus 4.8 on its internal organic-chemistry benchmark for inferring molecular structures from spectroscopy data, and 7.7 points higher on predicting how protein sequence variations affect function.
Two quieter platform changes worth more than the model card
Two developer-facing releases shipped alongside the model, and for teams running Claude in production they may matter as much as the benchmark deltas. First, the Claude Platform now supports changing which tools a model can use mid-conversation without invalidating the prompt cache — a real cost fix for agentic workloads, where cache invalidation on tool swaps quietly inflates bills. Second, the API gains automatic fallbacks: requests flagged by Anthropic’s safety classifiers on Opus 5 or Fable 5 can now route to another model instead of being blocked outright, so requests always land on the best available model by default.
Add it up and the strategy is legible: Anthropic is compressing the price of near-frontier intelligence on a two-month clock, while keeping one genuinely scarce capability — Fable 5’s ceiling — as the premium tier. For most teams, the practical question this week is no longer whether to upgrade from Opus 4.8, which now looks strictly dominated at identical pricing, but whether Fable 5 still justifies its premium for anything outside the hardest problems.
