Claude Opus 5.5 is now available across Anthropic’s platforms, Amazon Web Services, Google Cloud and Microsoft Azure, giving developers a new top-tier model for long, multi-step work at lower per-token prices. Early tests show the model completing a 200,000-line codebase audit in under three hours, down from more than 20 hours on the prior version, and shipping with new safeguards against prompt injection and model distillation.
What Claude Opus 5.5 is built to do
Opus 5.5 is designed for the kind of work that runs unattended for hours: codebase migrations, software audits, financial analysis, data collection and workflows that span several applications. The model can be accessed through the Claude Platform using the model name claude-opus-5-5.
In one benchmark test, Opus 5.5 audited and fixed a 200,000-line codebase in less than three hours. Opus 5 took more than 20 hours on the same task and used 2.5 times as many tokens to finish. Testers reported that the new model held context longer, delegated work to other agents more reliably and checked its own results before returning them.
How it performs on professional benchmarks
Anthropic reports gains in coding, computer use and professional knowledge work. On Terminal-Bench 4.0, a benchmark that measures how well a model can complete complex, multi-step professional tasks inside a command line interface, Opus 5.5 outperformed Opus 5 across the tested configurations.
Independent early users highlighted similar patterns. At the lowest effort setting, the model caught 72% of known bugs in code reviews against 56% for Opus 5 at high effort, with fewer false alarms and a fraction of the output. On US consulting analysis, low thinking effort matched higher thinking effort on half the output and still passed quality checks, suggesting more lower-effort runs can move into production.
Performance varies based on the task, available tools, and developer-configured reasoning effort.
Thinking mode and preserved context
Opus 5.5 cannot run with thinking switched off. The model always uses a reasoning process, and developers can tune how much effort it applies to a given task. This is a departure from the optional reasoning toggles some users had on prior Claude versions.
The release also introduces preserved thinking, a safeguard aimed at API users. Preserved thinking makes it harder for API users to edit Claude’s prior context, which Anthropic says reduces the risk that someone could extract the model’s capabilities through large-scale distillation attacks. Preserved thinking applies to Opus 5.5 API accounts created on or after August 31, 2026.
New safeguards for sensitive work
Opus 5.5 ships with three categories of new safety controls covering cybersecurity, biology and attempts to copy the model.
- Most cybersecurity tasks are rerouted to Opus 4.8. Anthropic plans to expand its Cyber Verification Program so verified cybersecurity professionals get broader access to Opus 5.5.
- Organizations whose biological research is impeded by the safeguards can apply to Anthropic’s Life Sciences Verification Program.
- A new classifier screens coding-agent actions before execution. Anthropic has also released an open-source sandbox that security teams can audit, plus code-review features designed to catch vulnerabilities before changes are merged.
Defenses against prompt-injection attacks have been strengthened. In an evaluation of attempts to cross containment boundaries, the model tried to circumvent its assigned limits about 85% less often than Opus 5 or Claude Mythos 5.1. Every attempt was classified as low severity and self-reported.
External organizations, including METR and Frontier Design, evaluated the model before release. Anthropic notes that current evaluations cannot identify every potential failure before deployment, and that risk assessments will need to keep pace as capability increases.
API pricing and throughput changes
Token prices for Opus 5.5 are lower across the board:
- Input: $4 per million tokens, down from $5.
- Output: $20 per million tokens, down from $25.
- Cached input reads: $0.20 per million tokens, down from $0.50.
The lower cache price particularly benefits coding agents and other systems that consult the same instructions, files or conversation history repeatedly, since cached reads are the dominant cost in those workloads.
Standard output is more than 30% faster than Opus 5. A separate fast mode is available through Claude Code and the Claude Platform, delivering up to 2.5 times the speed at $8 per million input tokens and $40 per million output tokens.
Higher usage limits for subscribers
Anthropic is raising five-hour usage limits for Pro, Max, Team and seat-based Enterprise customers. Subscription users will also receive a rate-limit reset they can save and use later, giving them a way to bank capacity for a burst job rather than losing unused quota at the end of a window.
For teams running long-running agents, lower token costs, higher sustained throughput, and expanded rate limits make it cheaper to leave an agent working overnight on a multi-repository task.
FAQ
What is Claude Opus 5.5?
Claude Opus 5.5 is Anthropic’s newest flagship model, designed for long, complex tasks such as codebase migrations, software audits, financial analysis and multi-application workflows. It is available on the Claude Platform, Amazon Web Services, Google Cloud and Microsoft Azure.
How much cheaper is Claude Opus 5.5 than Opus 5?
Input tokens cost $4 per million (down from $5), output tokens cost $20 per million (down from $25), and cached input reads cost $0.20 per million (down from $0.50). Output is also more than 30% faster.
What new safeguards does Claude Opus 5.5 include?
The model adds preserved thinking to resist distillation attacks, a classifier that screens coding-agent actions before execution, an open-source security sandbox, stronger prompt-injection defenses and verification programs for cybersecurity and life sciences work. External evaluators from METR and Frontier Design reviewed the model before release.
This article summarizes reporting from helpnetsecurity.com.