Anthropic released Claude Opus 5 on Friday, positioning it as the everyday AI model for enterprises and developers who need high performance without the premium price tag of Fable 5. The new model matches or exceeds Fable 5's performance on most benchmarks while costing half as much, with pricing remaining at $5 per million input tokens and $25 per million output tokens, unchanged from the previous Opus 4.8 version.

The release marks a significant shift in Anthropic's model lineup. While Fable 5 remains the company's most capable model for ambitious, long-running projects requiring days of autonomous work, Opus 5 is designed for the tasks that occupy most developers' and knowledge workers' daily routines. It becomes the default model for Claude Max subscribers and the best option available to Claude Pro subscribers.

How Does Opus 5 Compare to Other Top AI Models?

On nearly every benchmark Anthropic has published, Opus 5 outperforms Fable 5, a surprising result that underscores the complexity of modern AI model development. The performance gaps are particularly striking in specific domains:

• Knowledge Work: Opus 5 scored 1,861 on the GDPval-AA v2 knowledge work benchmark, ahead of Fable 5 (1,747), OpenAI's GPT-5.6 Sol (1,736), and Opus 4.8 (1,593).
• Business Workflows: Opus 5 achieved the top score on Zapier's AutomationBench, which measures AI agents on end-to-end business tasks, with a pass rate double that of the next-best model at the same cost per task.
• Coding Performance: On CursorBench 3.2, Opus 5 beats every other model on performance-per-cost at high, extra high, and max settings, landing within 0.5 percent of Fable 5's peak score at half the cost per task.
• Reasoning Tasks: Opus 5 edges out Fable 5 on Humanity's Last Exam with tools access, scoring 64.7 percent versus Fable 5's 63.9 percent.

The model also demonstrates improvements in autonomous work capabilities. Anthropic says Opus 5 can autonomously perform tasks for much longer than before, checking its own work and recovering from errors without human intervention.

What Makes Opus 5 Different from Previous Versions?

Opus 5 represents the latest iteration in a rapid development cycle. Opus 4 launched in May 2025, followed by Opus 4.5 in November 2025, Opus 4.6 in February 2026, Opus 4.7 in April 2026, and Opus 4.8 just a month and a half before this release. The frequent updates reflect Anthropic's strategy of continuous improvement across its model family.

The new model is particularly strong for knowledge work and coding tasks. Harvey, a legal AI company, reported its biggest gains with Opus 5 in practice areas like corporate governance and arbitration. The company noted that Opus 5 maintains quality at lower reasoning levels, achieving similar performance while generating 26 percent fewer tokens on average compared to Opus 4.8 at maximum reasoning effort.

Like Opus 4.8, Opus 5 offers a "fast mode" that generates output tokens 2.5 times faster, though at double the cost. This flexibility allows users to trade speed for economy depending on their needs.

How Is Anthropic Handling Safety and Cybersecurity?

Anthropic intentionally did not train Opus 5 on cybersecurity tasks, yet the model still comes close to Mythos 5, Anthropic's specialized health model, at finding vulnerabilities in open-source code. Critically, it remains far behind at actually exploiting those vulnerabilities, which is exactly where Anthropic wants the capability balance to sit.

To maintain this safety boundary, Opus 5 ships with safety classifiers that screen requests for a narrow range of cyber tasks, including exploit generation and penetration testing. Because these classifiers are scoped more narrowly than Fable 5's restrictions, Anthropic expects them to intervene about 85 percent less often, leaving room for defensive security work while blocking techniques more likely to be associated with malicious actors.

Anthropic also calls Opus 5 its most aligned model to date based on automated behavioral audits, and together with Fable 5, the least susceptible to being tricked into misuse. When a safety classifier does flag a request in Claude.ai, Claude Code, or Cowork, the system falls back to Opus 4.8 by default. API users can enable the same behavior, and enterprises participating in Anthropic's Cyber Verification Program get access to a version with fewer restrictions.

What Do These Updates Mean for Developers and Enterprises?

Cursor, the AI-powered code editor, is already integrating Opus 5. Sualeh Asif, Cursor's co-founder, stated that "Claude Opus 5 delivers near Fable 5 intelligence at Opus speed and cost. On CursorBench it's just under Fable 5 and has many of the same behaviors. We are excited to see how developers use it in Cursor".

Sualeh Asif, Cursor's co-founder

"Claude Opus 5 delivers near Fable 5 intelligence at Opus speed and cost. On CursorBench it's just under Fable 5 and has many of the same behaviors. We are excited to see how developers use it in Cursor," said Sualeh Asif, Cursor co-founder.

Sualeh Asif, Co-founder at Cursor

The release also introduces two smaller platform updates in beta. Automatic fallbacks allow API requests that Anthropic's safety classifiers decline on Opus 5 or Fable 5 to automatically route to another model within the same request instead of returning an error. Developers can also now change which tools Claude can use mid-conversation without invalidating the prompt cache, so each phase of an agent's work only sees the tools it needs.

Opus 5 is available today on all of Anthropic's paid plans and on the Claude Platform as claude-opus-5. The pricing strategy reflects what Anthropic sees as a critical market concern: every business is now sensitive to token costs, and every AI lab is acutely aware of how its pricing is perceived relative to competitors.