Anthropic PBC introduced Claude Opus 5 on Friday, presenting the model as a cost-efficient option for business-oriented artificial intelligence tasks at a time when customer price sensitivity is rising and competition in China is intensifying.
The company says Claude Opus 5 attains performance that in many respects approaches its most advanced offering, Fable 5, while reducing the cost per task by roughly half. Anthropic expects the new release to become the standard choice for a range of routine office functions.
Pricing for Opus 5 is set at $5 per million input tokens and $25 per million output tokens, the same unit rates Anthropic applies to Opus 4.8, its predecessor. Despite keeping the same base token prices, Opus 5 is advertised as delivering improved performance compared with Opus 4.8. The company said the model is available across all supported platforms beginning Friday.
Anthropic highlighted several benchmark results to illustrate Opus 5's capabilities. On software engineering assessments, Opus 5 led on Frontier-Bench v0.1, outperforming other models and achieving more than double the performance of Opus 4.8 while reducing the cost per task. On CursorBench 3.2, Opus 5 reached a score within 0.5% of Fable 5’s top score when run at maximum effort, yet did so at roughly half the cost per task.
For problem-solving challenges, the company reported that Opus 5 achieved a score on ARC-AGI 3 that was three times higher than the next-best model. In end-to-end business automation testing on Zapier AutomationBench, Opus 5 showed a pass rate about 1.5 times that of the next-best model when compared at the same cost per task.
Anthropic also cited improvements in scientific research tasks. The new model scored 10.2 percentage points higher than Opus 4.8 on organic chemistry tasks and 7.7 percentage points higher on protein-related tasks, according to the company’s results.
On safety metrics, Opus 5 recorded an overall misaligned behavior score of 2.3, the lowest among the company’s recent models. Anthropic said Opus 5 does not advance risky, dual-use capabilities and noted that it still trails Mythos 5 in areas of biology research and offensive cybersecurity.
Operationally, Anthropic has made Opus 5 the default model on its Claude Max tier and identifies it as the strongest available model on Claude Pro. The company also offers a Fast mode for Opus 5 that runs at about 2.5 times the default speed; that mode is priced at double the base rate.
Implications
- Anthropic presents Opus 5 as a lower-cost workhorse for companies focused on routine office and software engineering tasks.
- Performance gains are claimed across software engineering, problem-solving, business automation, and certain scientific tasks versus Opus 4.8.
- Safety testing shows reduced misalignment scores compared with recent Anthropic models, though some specialized research capabilities remain stronger in other Anthropic models.