Anthropic shipped Claude Opus 5, its fourth model release in under two months following Mythos 5, Fable 5, and Sonnet 5, and positioned it as a model that gets close to Fable 5-level intelligence while costing roughly half as much to run. On coding and knowledge-work evaluations like Frontier-Bench and GDPval-AA it's described as state of the art for Anthropic, though it reportedly still trails Mythos 5 specifically on cybersecurity-focused tasks. It's now the default model on Claude Max and the strongest option available on Claude Pro, and on third-party leaderboards it reportedly edges out both Fable 5 and GPT-5.6 Sol on rebased intelligence and agentic-task indices. The detail most relevant to people actually building on top of it is the effort control: Opus 5 exposes a per-request toggle for how much reasoning effort the model should spend, low, medium, or high, letting a developer trade latency and cost against answer quality on a per-call basis rather than being locked into whichever reasoning depth the model defaults to. This mirrors a broader pattern spreading across the industry right now, OpenAI's GPT-5.4 and GPT-5.6 APIs and LangChain's core library both shipped similar effort or reasoning-budget parameters within days of each other this month, suggesting that exposing reasoning depth as a first-class, tunable parameter is becoming a standard expectation for frontier model APIs rather than a one-off feature. For teams already building on Claude, the practical upshot is a cost and quality lever worth wiring into any latency-sensitive or budget-sensitive workflow, cheap, low-effort calls for routine work, dialed up only when a task actually needs deeper reasoning, all through the same model rather than switching model tiers.