Anthropic launches Claude Opus 5, claims near-Fable 5 performance at half the token price

Anthropic released Claude Opus 5 on July 24, 2026, positioning it as a cheaper alternative to its Fable 5 model. The new flagship is available on Amazon Bedrock and Anthropic's own platform, including Claude Pro and Claude Max tiers.

Opus 5 costs $5 per million input tokens and $25 per million output tokens, half of Fable 5's rates. The company says it delivers near-Fable 5 performance on several benchmarks, while its safety classifiers intervene around 85% less often than those on the Fable model.

Pricing and performance benchmarks

On Frontier-Bench v0.1, a test of agentic terminal coding, Opus 5 scored 43.3%, beating Fable 5 (33.7%) and GPT-5.6 Sol (34.4%). On the knowledge work benchmark GDPval-AA v2, Opus 5 achieved an Elo score of 1,861, again ahead of Fable 5 (1,747) and GPT-5.6 Sol (1,736).

• On ARC-AGI-3, which measures novel problem-solving, Opus 5 hit 30.2%, nearly four times higher than GPT-5.6 Sol (7.8%).
• On the Artificial Analysis Coding Agent Index, Opus 5 reached the highest overall score, narrowly ahead of GPT-5.6 Sol.
• On DeepSWE v1.1 agentic coding, GPT-5.6 Sol led at 72.7%, followed by Fable 5 (69.7%) and Opus 5 (68.8%).
• Opus 5 offers five effort settings (low, medium, high, xhigh, max). At the max setting, performance dropped on two benchmarks despite higher cost.
• A new Fast Mode boosts speed by 2.5x but doubles the price.

Opus 5 also improves at generating visual outputs and analyzing charts and diagrams. The model can build its own tools through code when needed, such as writing a computer vision pipeline to extract geometry from raw pixels.

Safety and availability

Anthropic says Opus 5 is the least susceptible to being tricked into misuse among its models. The company deliberately did not train Opus 5 on cybersecurity tasks, resulting in worse exploit development performance compared to its Mythos 5 model. However, Opus 5 finds vulnerabilities about as well as Mythos 5.

Opus 5 is Anthropic's fourth model release in less than two months, following Fable 5, Mythos 5, and Sonnet 5. The model is excluded from Anthropic's 30-day data retention policy, meaning user data is not stored for retraining. Enterprises on AWS Bedrock can integrate Opus 5 into agentic systems and production inference workloads immediately.

With pricing pressure from GPT-5.6 Sol and Chinese competitors, Anthropic is betting that lower token costs and strong benchmark scores will attract developers who found Fable 5 too expensive. The real test will be whether Opus 5's efficiency gains hold up in production workloads, where token usage can vary significantly.