Anthropic introduced Claude Opus 5 on July 24, 2026, positioning it to deliver intelligence comparable to Fable 5 but at just half the price. This move targets developers and enterprises eager for top-tier coding and agent performance without the hefty fees typically associated with flagship AI models.
Following up on Opus 4.8 released two months earlier, Opus 5 maintains the same operating cost yet significantly improves performance. The model excels on the Frontier-Bench v0.1 coding assessment, outperforming all competitors and doubling Opus 4.8’s effectiveness per task. Users can adjust an effort setting to balance between peak intelligence or quicker, more cost-efficient responses.
Opus 5 is now the default for Claude Max and the prime option for Claude Pro, designed as a versatile everyday workhorse rather than a niche specialist. On CursorBench 3.2, it scores within half a percentage point of Fable 5’s top results, but at half the cost per task. Impressively, it surpasses Fable 5 on several benchmarks. For instance, on ARC-AGI 3’s novel problem test, Opus 5 performed three times better than the next best model, and on Zapier’s AutomationBench, its pass rate was about 1.5 times higher than the nearest competitor, all while keeping costs low. It also outdid Fable 5 in OSWorld 2.0’s computer use benchmark for a fraction of the price.
One tradeoff is that Opus 5 carries fewer safety restrictions than Fable 5 but introduces a new Automatic Fallbacks feature. Instead of returning errors when requests are blocked, the system reroutes them, aiming to improve user experience. However, Anthropic acknowledged that Opus 5 falls short in cybersecurity tasks compared to Mythos 5.
Unlike Fable and Mythos, Opus 5 is exempt from Anthropic’s 30-day data retention policy, a factor that may influence enterprise adoption decisions.
