Claude Opus 5 Approaches Fable 5 With Adjustable Reasoning Effort, API Pricing Remains $5 per Million Input Tokens

Anthropic launched Claude Opus 5 on July 24, positioning the product between the top-tier Fable 5 and lighter-weight models. API pricing is $5 per million input tokens and $25 per million output tokens, unchanged from Opus 4.8 and half the price of Fable 5. Developers can also adjust `effort`, allowing the same model to trade off reasoning depth, latency, and token consumption based on the task. This is more straightforward than maintaining separate routing rules for multiple models, but whether low `effort` can consistently preserve quality on complex tasks must still be validated workload by workload.

GitHub has added Opus 5 to Copilot Pro+, Max, Business, and Enterprise across VS Code, JetBrains IDEs, the CLI, cloud agents, and mobile. GitHub says its early testing focused on autonomous code changes, regression validation, and multi-tool coordination, and found that the model was better able to target necessary changes, verify results, and reduce unnecessary execution. These claims have not yet been accompanied by public test sets, complete trajectories, or cost distributions, so they should not be treated as independent benchmarks.

On safety, Opus 5 strengthens safeguards against high-risk cybersecurity content, which may also block legitimate vulnerability research or security-adjacent requests. In addition to comparing task success rates, engineering teams should track refusal rates, tool-call counts, end-to-end costs, and interruption rates during long-running tasks. Fable 5 remains positioned as the preferred option for the most complex, long-running autonomous work. Opus 5’s real value therefore depends on whether it can maintain a comparable completion rate at lower cost across real-world agent trajectories—not on one-off benchmark scores.