Anthropic's Opus 5 is about token efficiency, not a capability leap

Anthropic has released Opus 5, an update to its AI model, showing iterative performance improvements in benchmarks for coding tasks, matching or slightly exceeding Opus 4.8 and OpenAI’s GPT-5.6-Sol. The primary pitch for Opus 5 is its cost efficiency, offering performance comparable to Anthropic’s Fable model at approximately half the price, though it deliberately lags in cybersecurity exploitation due to training decisions.

• Opus 5 represents an iterative performance increase for coding tasks, not a breakthrough.
• It performs similarly to or slightly better than Anthropic’s Fable model on benchmarks like Frontier-Bench and DeepSWE.
• Opus 5 is positioned as a more cost-effective alternative to Fable, costing half as much.
• The model has intentionally been trained to lag in cybersecurity exploitation tasks.
• The main focus for developers and managers is cost reduction, driving interest in open-weight and local models.
• Opus 5’s pricing is $5 per million input tokens and $25 per million output tokens.
• Competition is increasing, with models like China’s Kimi K3 offering similar performance at lower costs.
• Companies are developing ‘model routers’ to optimize AI model usage and reduce costs.
• Anthropic needs to continue reducing costs or increasing performance-for-cost to maintain growth.
  Continue reading https://arstechnica.com/ai/2026/07/anthropics-opus-5-is-about-token-efficiency-not-a-capability-leap/