Anthropic has launched Claude Opus 5, a new AI model with improved performance and lower costs, now available across all platforms.
Anthropic has introduced Claude Opus 5, a new AI model that reportedly brings performance levels close to its leading model, Fable 5. This latest version is now accessible across all platforms.
Designated as the default model for Max subscribers, Claude Opus 5 maintains the same pricing structure as its predecessor Opus 4.8, costing $5 per million input tokens and $25 per million output tokens.
According to the company, Opus 5 demonstrates performance comparable to Claude Fable 5 while significantly reducing token costs. It has established new benchmarks in programming and cognitive assessments, such as Frontier-Bench and GDPval-AA, though it still falls short of Mythos 5 in cybersecurity capabilities.
In the Frontier-Bench v0.1 benchmark, Opus 5 more than doubled the performance of Opus 4.8 at a lower cost per task. It also scored nearly on par with Fable 5 in CursorBench 3.2, achieving a result just under 0.5% less than the peak score while costing half as much.
In the ARC-AGI 3 test, Opus 5 surpassed its closest rival by a factor of three. Its success rate on the Zapier AutomationBench platform was around 1.5 times higher for the same financial investment. Moreover, in the OSWorld 2.0 test, it outperformed Fable 5's best result.
The new model has shown significant improvements in scientific evaluations as well, outperforming Opus 4.8 in all life sciences assessments. Notably, it achieved a 10.2 percentage point improvement in organic chemistry and a 7.7 percentage point increase in protein-related tasks.
Anthropic claims that Opus 5 is the most compliant model developed so far, registering the lowest inconsistency score of 2.3 in an automated behavioral audit among its recent models.
In terms of cybersecurity, the company has eased restrictions compared to Fable 5, predicting that classifier interventions will occur about 85% less frequently. Requests flagged by security systems will be automatically rerouted from Claude.ai, Claude Code, and Claude Cowork to Opus 4.8, with a similar option available through the API.
Furthermore, biological queries previously blocked on Fable 5 will now be processed by Opus 5 instead of Opus 4.8.
Along with the launch of Opus 5, Anthropic has rolled out two beta features: users can now change available tools during a conversation without resetting the prompt cache, and there are automatic switches to backup models in the API.
It is noteworthy that Opus 4.8 was released in May, alongside the introduction of dynamic workflow features in Claude Code.
