Anthropic Releases "Claude Opus 5"

Anthropic has unveiled its new AI model "Claude Opus 5." According to the company, this model delivers performance close to its flagship model "Claude Fable 5" while cutting costs to roughly half. Pricing is set at $5 per million input tokens and $25 per output token, maintaining the same rates as the previous generation Opus 4.8. The model became available across all Anthropic platforms upon release and has been set as the default model for the premium personal plan "Claude Max."

The competition in AI has traditionally centered on "how intelligent" each model is—a pure performance comparison. However, as enterprises integrate AI into their actual workflows, the evaluation axis is increasingly shifting toward "economic rationality"—whether equivalent quality can be delivered at lower cost and broader accessibility. Anthropic's positioning of Opus 5 as a "cost-efficient practical model" can be seen as a response to this market shift. Rather than competing on cutting-edge performance, the company is asserting dominance in the bandwidth of tasks that enterprises most frequently handle in daily operations—those that are "neither too difficult nor too easy."

Anthropic has organized its model lineup into four tiers. Fable 5 is designed for large-scale autonomous projects operating on a multi-day timescale, Opus 5 for complex everyday work such as coding and knowledge work, Sonnet 5 for high-volume processing where speed and cost take priority, and Haiku 4.5 for sub-agent deployment and immediate response use cases. This structure clarifies a taxonomy for selecting models based on application.

Benchmark results demonstrate Opus 5's standout performance across several metrics. In "Frontier-Bench v0.1," which evaluates AI agents performing coding on a terminal, Opus 5 achieved a score of 43.3%—more than double the previous-generation Opus 4.8 (18.7%) and surpassing Fable 5 (33.7%). In "OSWorld 2.0," which assesses PC automation capabilities, Opus 5 exceeded Fable 5's best score while achieving this at roughly one-third the cost of Fable 5. On "ARC-AGI 3," which measures novel problem-solving ability, Anthropic reports Opus 5 achieved three times the score of the next-best model.

Notably, Anthropic itself has been candid about the model's limitations. The company has acknowledged underperformance relative to competing models in cybersecurity tasks and biological research domains, and that other companies' models lead in certain agent-based coding evaluations. Additionally, Anthropic has stated that benchmarks excel with "finite tasks with predetermined outcomes" but are poorly suited for evaluating sustained work over extended periods, suggesting that scores capture only part of the picture.

This transparency can be interpreted as a response to growing skepticism toward benchmark competition within the AI industry. By prioritizing honest communication about real-world applicability over numerical boasting, Anthropic is positioning itself to earn enterprise customer trust. In corporate deployment scenarios, processing efficiency for routine daily work—rather than extreme edge cases—often drives decision-making, and Opus 5's value proposition aligns precisely with this reality.

The evolution of AI models is entering a phase in which the evaluation axis is shifting from simple "performance excellence" to "balance between cost and performance." What Anthropic has demonstrated is a model design philosophy that achieves both cutting-edge capability and practical cost efficiency. The coming focal point will be how widely this design philosophy is embraced in the future enterprise AI market.