Anthropic's Opus 5 focuses on efficiency, not capability leap

AI company Anthropic has rolled out Opus 5, the latest update to its language model that has gained popularity for tasks including coding and software development.

While Opus 5 represents an improvement, it does not appear to be a breakthrough in agentic coding performance comparable to previous updates. Benchmarks show Opus 5 performing at or slightly above Anthropic's Fable model for coding, while also outperforming Opus 4.8 and OpenAI's GPT-5.6-Sol across various tasks.

The company is highlighting the model's cost-effectiveness. Opus 5 offers performance close to Fable at approximately half the cost. Anthropic has intentionally limited Opus 5's cybersecurity training, positioning it behind specialized models like Fable and Mythos in exploit capabilities. It is priced at $5 per million input tokens and $25 per million output tokens, matching its predecessor but undercutting Fable.

This release comes amid a market focus on cost optimization. Companies are developing "model routers" to select the most efficient AI model for specific tasks, aiming to save on token usage and associated costs. Intense competition exists, with models like China's Kimi K3 offering similar performance at a lower price point.

To maintain growth, Anthropic and similar companies must continue to reduce token costs or offer increased performance for the same price. Failure to do so risks them losing market share to smaller, open-source, or more cost-effective alternatives as they improve.