Anthropic released Claude Opus 5 on July 24, 2026, offering performance close to its frontier Fable 5 model at half the cost. The new model targets developers and enterprises with enhanced coding and reasoning capabilities.
Claude Opus 5, Anthropic's latest flagship model, is now available, providing performance comparable to the company's top-tier Fable 5 model but at a significantly reduced price. This release aims to make advanced AI capabilities more accessible to businesses and developers, featuring improved efficiency in coding tasks and complex reasoning. Opus 5 replaces Opus 4.8 as the default model for Claude Max and is the strongest option for Claude Pro users.
The new Opus 5 model demonstrates substantial gains across various benchmarks, particularly in coding and knowledge work evaluations. On the Frontier-Bench v0.1, Opus 5 more than doubled the performance of its predecessor, Opus 4.8, and surpassed all other models tested. In coding evaluations such as CursorBench 3.2, Opus 5 achieved performance within 0.5% of Fable 5's peak score while operating at half the cost per task. Anthropic highlighted that Opus 5 excels even at its lowest effort setting, outperforming other models and completing more tasks at a given cost.
For general reasoning and problem-solving, Opus 5 also shows marked improvements. On the ARC-AGI 3 benchmark, designed to test a model's ability to solve novel problems, Opus 5 scored three times higher than the next best model. Similarly, Zapier AutomationBench, which measures the completion of business tasks, saw Opus 5 achieve a 1.5 times higher pass rate for the same cost per task compared to other models. On OSWorld 2.0, a benchmark for computer use, Opus 5 outperformed all other models at any given cost, exceeding Fable 5's best result at just over a third of the expense.
Anthropic has maintained the pricing structure for Opus 5, keeping it at $5 per million input tokens and $25 per million output tokens, the same as Opus 4.8. This pricing strategy, combined with its enhanced performance, positions Opus 5 as a more cost-effective option for demanding AI applications. The company also offers a "Fast mode" for Opus 5, which runs approximately 2.5 times faster than the default setting, available at double the base price on the Claude Platform and through usage credits in Claude Code.
The model features a 1 million token context window and a maximum output of 128,000 tokens. Its knowledge cutoff is May 2026, making it the most current among Anthropic's Claude models. Opus 5 also supports zero data retention, a feature not available with Fable 5, which requires a 30-day data retention period.
Anthropic noted that Opus 5 is designed for everyday use and operates more efficiently than previous models. Developers integrating Opus 5 into their applications should be aware that "thinking" is now enabled by default, and instructions for adding explicit verification steps may cause over-verification, as the model inherently verifies its own work. While Opus 5 sets new benchmarks in many areas, it remains behind Mythos 5 in cybersecurity tasks.