Anthropic releases Claude Opus 5.5 with 40% lower costs

Anthropic has officially launched Claude Opus 5.5, the first model in its new 5.5 family, marking a significant shift in the economics of high-end AI coding. According to the release notes on Anthropic's website, the model delivers performance comparable to their top-tier Claude Fable 5.1 on most workloads but costs 40% less to run than the previous Opus 5. This release is particularly notable as it is the first major model update since Anthropic called for pacing the development of frontier AI, with the company stating that the model was tested by external evaluators including Frontier Design and METR before release.

Performance and Efficiency Gains

The primary selling point of Opus 5.5 is its ability to handle long, sprawling engineering jobs with significantly higher efficiency than its predecessor. Anthropic reports that an early tester completed a 680,000-line code migration in less than a day, work that would have taken an engineering team weeks. In another internal test, Opus 5.5 translated the HAProxy software from C to Rust in 9.5 hours, compared to 12 hours for Claude Fable 5.1, while costing 51% less.

The model also shows marked improvements in reliability for specific tasks. When asked to cut load times across every page of a web app, Opus 5.5 succeeded in 39 out of 40 attempts, whereas Opus 5 made smaller improvements that often altered the app’s behavior in unintended ways. On the FrontierCode benchmark, which measures whether an agent’s code changes would be merged, Opus 5.5 scored 54.6% at default effort, beating GPT-6 Astra’s top score of 53.3% for roughly one-fifth of the cost per task.

Pricing and Speed Improvements

Anthropic has adjusted pricing to reflect the reduced compute requirements of the new model. Input tokens are priced at $4 per million and output tokens at $20 per million, a 20% reduction from Opus 5. More significantly for agentic and coding workflows, cache reads— which make up the majority of costs in these scenarios—are now $0.20 per million tokens, a 60% decrease. The model also generates output more than 30% faster than Opus 5.

For developers using the Claude Code platform, a "fast mode" is available with up to 2.5x speed, priced at $8 per million input tokens and $40 per million output tokens. Anthropic has also increased five-hour usage limits on Pro, Max, Team, and seat-based Enterprise plans, and introduced a rate limit reset feature that allows subscription users to save and use their limits at their discretion.

Safety and Alignment

Anthropic claims Opus 5.5 achieves the best scores of any model to date on their automated behavioral audit, an alignment suite that tests Claude across thousands of simulated scenarios. The model is described as being much less likely than recent models to take hard-to-reverse actions or act outside given boundaries. It is also more resistant to prompt injection than Opus 5.

Due to its capabilities in biology and cybersecurity, comparable to Claude Mythos 5.1, the model is deployed with safeguards similar to those on Claude Fable 5.1. Vetted organizations can apply to the Life Sciences Verification Program to use Opus 5.5 for biology research, and access to the Cyber Verification Program is expected to expand in the coming weeks for verified cybersecurity practitioners.

What It Means for Developers

For developers, the shift to Opus 5.5 represents a tangible reduction in the cost of running complex agentic workflows. The 40% cost reduction, combined with faster output generation, makes it more viable to use frontier-level models for large-scale codebase audits and migrations without the prohibitive expense previously associated with such tasks. The improved reliability in multi-step tasks, such as the HAProxy translation and web app optimization tests, suggests that the model is better suited for production environments where consistency is critical.

However, developers should be aware of the safeguards in place. When safeguards intervene, cybersecurity tasks are completed by Claude Opus 4.8, and biology and frontier LLM development tasks are completed by Claude Opus 5. This may reduce performance on specific benchmarks but reflects Anthropic's commitment to safety. For teams working on sensitive or high-stakes projects, the availability of verification programs provides a pathway to access the full capabilities of the model while adhering to safety guidelines.

What to Watch

• Release of Sonnet 5.5 and Haiku 5.5: Anthropic has announced that these models will follow in the coming weeks, likely bringing similar improvements to performance, efficiency, and safety to the mid-tier and lightweight segments.
• Expansion of Verification Programs: The rollout of the Cyber Verification Program and the expansion of the Life Sciences Verification Program will determine how widely the model's full capabilities can be accessed by verified practitioners.
• Real-World Benchmark Discrepancies: Anthropic notes that benchmark margins have become less reliable guides to real-world differences at these capability levels. Developers should monitor real-world performance in their specific workflows, as the gap between Opus 5.5 and Claude Fable 5.1 may be narrower than scores suggest.
• External Evaluator Feedback: As the first release since the call for pacing the frontier, the feedback from external evaluators like Frontier Design and METR will be crucial in assessing the model's safety and alignment in independent settings.