Claude Opus 5 puts near-Fable coding power at half the cost

The most important number in Claude Opus 5 is not a raw leaderboard score. It is the price-performance claim: Anthropic says the model gets close to Claude Fable 5 on coding and knowledge work at half the cost. Opus 5 is available now, becomes the default model on Claude Max, and is positioned as the strongest option for Claude Pro users. API pricing stays at the Opus 4.8 level: $5 per million input tokens and $25 per million output tokens.

That framing matters because frontier-model adoption is increasingly constrained by unit economics, not only capability. Anthropic says Opus 5 leads on Frontier-Bench v0.1 and more than doubles Opus 4.8’s performance at a lower cost per task. On CursorBench 3.2 at max effort, the model reportedly comes within 0.5% of Fable 5’s peak score while costing half as much per task. The company also cites a threefold lead over the next-best model on ARC-AGI 3 and roughly 1.5x the next-best pass rate on Zapier AutomationBench at the same cost per task.

The practical claim is about longer agent runs. Anthropic describes Opus 5 as stronger at checking its own work, iterating, and carrying complex tasks to completion. Its examples include a Frontier-Bench task where the model, denied direct access to a drawing, built a computer-vision pipeline from raw pixels and reconstructed a 3D FreeCAD part. In another example, it found the root cause of a real bug in an open-source package manager and fixed an edge case missed by a community patch. Those are cherry-picked examples, but they point to the product bet: fewer shallow fixes, more finished work.

Opus 5 also gets a science pitch. Anthropic says it beats Opus 4.8 across its life-sciences evaluations, including structural biology, organic chemistry, and bioinformatics. The largest named gains are 10.2 percentage points on internal organic-chemistry tasks that infer molecular structures from spectroscopy data, and 7.7 percentage points on protein tasks that predict how sequence variations affect function. For research teams, those task-specific deltas are more useful than a generic chat score because they map to workflows where mistakes are expensive to investigate.

The safety section is just as central to the launch. Anthropic says Opus 5 does not push the frontier in risky dual-use capabilities and remains behind Mythos 5 in biology research and offensive cybersecurity. Its safeguards resemble Opus 4.8’s, with stronger guardrails for a narrow set of cyber tasks. At the same time, the company says Opus 5’s cyber classifiers should intervene around 85% less often than Fable 5’s. Flagged requests in Claude.ai, Claude Code, and Claude Cowork can fall back to Opus 4.8 by default, and API customers are getting beta support for automatic fallbacks.

The competitive question is whether Opus 5 changes default routing. Anthropic is leaving Fable 5 as the higher frontier point while making Opus 5 the everyday high-end model. That lets teams split workloads by cost, effort setting, and risk controls instead of choosing one model for everything. The next signal to watch is not only benchmark movement, but how quickly coding tools, enterprise agents, and research assistants move routine Fable-class workloads onto Opus 5.