Claude Opus 5 promises near Fable performance at half the cost

Anthropic has introduced Claude Opus 5, a new AI model positioned as a lower-cost alternative to its most capable commercially available system, Claude Fable 5. The company says Opus 5 approaches Fable 5’s performance on coding, knowledge work and computer-use tasks while costing roughly half as much per task.

Opus 5 is now the default model for Claude Max subscribers and the most capable option on Claude Pro. Developers can access it through Anthropic’s API under the name claude-opus-5. Pricing remains unchanged from the previous Opus 4.8 model: $5 per million input tokens and $25 per million output tokens.

The launch reflects a broader shift in the AI market. Rather than presenting Opus 5 as the company’s absolute strongest model, Anthropic is targeting everyday professional workflows where price, speed and reliability matter as much as peak benchmark performance.

Built for coding, research and business automation

Anthropic says Opus 5 sets new company-leading results on several benchmarks for software development and knowledge work. On Frontier-Bench, which tests agentic coding tasks in a terminal environment, the company reports that Opus 5 more than doubles the score of Opus 4.8. It also claims the model delivers strong results on ARC-AGI 3, a test of novel problem solving, and OSWorld 2.0, which measures computer-use abilities.

For content and knowledge professionals, the more relevant claim may be efficiency. Opus 5 includes an adjustable effort setting. Users can choose lower effort for faster and cheaper answers, or higher effort when a task requires more detailed reasoning and checking.

Anthropic and early users say the model is better at reviewing its own work. In one example, the company says Opus 5 built a computer vision process to interpret the raw pixels of a drawing it could not directly view, then used the result to create a 3D CAD model. In another test, it reportedly found the underlying cause of a bug in an open-source package manager rather than merely correcting the visible symptom.

VentureBeat reports that Anthropic sees a distinction between bounded tasks and long-running autonomous work. Opus 5 is intended for complex but clearly defined assignments, while Fable 5 remains the company’s option for projects that require coherence across many hours or days.

Safety restrictions remain, but are narrower

Anthropic says its internal behavioral audit rates Opus 5 as its most aligned recent model. The company reports lower levels of deceptive behavior and a lower likelihood of taking reckless actions than Opus 4.8, Sonnet 5 or Fable 5.

The company says it did not specifically train Opus 5 on cybersecurity tasks. However, general improvements have made it better at identifying software vulnerabilities. Anthropic says the model remains substantially weaker than Mythos 5 at turning discovered vulnerabilities into working exploits.

Cybersecurity safeguards still block some requests, including penetration testing, exploit generation and binary-based vulnerability scanning. But Anthropic expects its classifiers to intervene about 85% less often than with Fable 5. When a request is flagged in Claude.ai, Claude Code or Claude Cowork, it can automatically fall back to Opus 4.8.

Anthropic is also introducing automatic fallback routing through its API and a beta feature that allows developers to change available tools during a conversation without losing the prompt cache. A Fast mode runs about 2.5 times faster than the standard version, at twice the base price.

Sources

• Introducing Claude Opus 5 – Anthropic
• Anthropic says Opus 5 can nearly match its top-performing model for half the price – Engadget
• Anthropic launches Claude Opus 5, a cheaper AI model for coding, agents and enterprise workflows – VentureBeat