Anthropic’s Claude Opus 5: High Performance at Half the Cost

Anthropic has launched Claude Opus 5, an AI model that edges close to its flagship Fable 5 while cutting costs in half. This new release shakes up the market with strong performance and lighter safeguards.

Opus 5 earns evaluation scores above 4.8 and outperforms Fable 5 on many benchmarks. It sets state-of-the-art records in agentic terminal coding, knowledge work, agentic search, and computer use tasks. The model scored a perfect 42 out of 42 on the International Math Olympiad 2026 problems, a rare feat in AI math reasoning.

Beyond raw scores, Opus 5 is the most aligned AI Anthropic has built. Its safeguards intervene 85 percent less than Fable 5, allowing more flexible use cases. These safeguards target beneficial applications in cybersecurity and biology, where openness matters. The model is also more resistant to deception and trickery, improving reliability in sensitive tasks.

Claude Opus 5 shines in complex coding and knowledge work, surpassing its predecessor, Opus 4.8. It achieves 90-95% accuracy on programming tasks and correctly handles keyboard and mouse operations 70% of the time. While it matches Mythos 5 in finding software bugs, it falls short in writing exploits, reflecting a prioritization of defensive security.

Pricing stays competitive: $5 per million input tokens and $25 per million output tokens. This matches Opus 4.8 and undercuts models like Kimi K3, which charges $15 per million output tokens despite offering Fable-like performance on some tasks. Anthropic claims Opus 5 may become the go-to model for work tasks, balancing cost, speed, and capability.

Accuracy varies by domain. Opus 5 answers general knowledge questions with a 22% error rate and health questions with 30%. Legal queries trip it up more, with a 70% error rate. These weaknesses underline its current limits despite advances.

Notably, Opus 5 is excluded from the 30-day data retention policy. This likely appeals to enterprises wary of data privacy rules. The model’s performance feels faster and more decisive than Fable 5, adding to its appeal for demanding users.

Opus 5’s combination of fewer restrictions, solid accuracy, and cost efficiency marks a clear step forward. It’s a model built for professionals who need power without the premium price and tight guardrails. Anthropic’s gamble on lighter safeguards aims to unlock more practical uses in cybersecurity and biology—fields that often require nuanced, flexible AI behavior.

This release doesn’t revolutionize AI capabilities, but it sharpens Anthropic’s lineup. Opus 5 improves token efficiency and task performance, carving out a niche between high-cost flagships and cheaper, less capable models. Expect it to become a staple for work-related AI tasks in the months ahead.

Stay connected via Google News

Follow us for the latest AI News