Anthropic's Claude Opus 5.5 tackles full coding workflows, but completion still doesn't mean correctness

Anthropic is pushing Claude into a broader role across the entire software development lifecycle. The company's new Claude Opus 5.5, the first release in a refreshed Claude 5.5 family, spans design specification creation, debugging, code generation and testing. According to Anthropic's benchmarks, Opus 5.5 matches the performance of Claude Fable 5.1 on most coding tasks while consuming roughly 40% less compute than Opus 5.

Mario Rodriguez, chief product officer at GitHub, endorsed the model's efficiency in a statement accompanying Anthropic's announcement. "In our testing [of Claude Opus 5.5] across GitHub Copilot CLI and VS Code, Claude Opus 5.5 used among the fewest tokens and steps we measured. In VS Code, it solved more terminal tasks than Opus 5 in less than half the steps. More than making individual tasks more efficient, it's making developers' bigger projects more achievable," Rodriguez said.

Safety testing and alignment practices

Anthropic developed the expanded capabilities of the Claude 5.5 family through what the company describes as established safety practices. The approach includes comprehensive alignment testing to ensure model outputs match human values and intended goals, pre-release evaluation by external organizations, and protective measures for high-risk domains including cybersecurity and biology.

On Anthropic's most comprehensive alignment test, Opus 5.5 achieved the strongest performance of any model tested to date, with notable improvements in behaviors linked to recent cybersecurity incidents, including biased reasoning and sandbox escape attempts.

Akash Thakur, an independent SRE and AI reliability architect, sees value in Anthropic's ability to make models complete entire coding tasks, but identifies a deeper challenge. "Models working at the level of Claude Opus 5.5 are genuinely good at breaking a project into small, finishable pieces — and that's the real unlock, because momentum on software comes from finishing things, not starting them," Thakur said. "…But 'completed' and 'correct' aren't the same thing. The task that looks done and completed is exactly the one that often costs the team later, so the win isn't removing the human — it's moving them from writing the code to verifying it."

Anthropic framed the release as evidence of a "higher bar for more capable models," arguing that systems capable of fully automating AI research itself should meet stricter safety standards. The company suggested that as AI capabilities expand, public policy should take on greater responsibility in ensuring these systems remain safe, pointing to recent work with Accenture as an example of infrastructure Anthropic is building to support this approach.

Large-scale codebase work

Anthropic highlighted Opus 5.5's particular strength on extended tasks like codebase-wide migrations and audits. An early tester reported that the model audited and corrected a 200,000-line codebase in under three hours, whereas Opus 5 required more than 20 hours and consumed 2.5 times as many tokens.

In an internal test, Anthropic asked both Opus 5.5 and Fable 5.1 to convert HAProxy, a widely deployed load-balancing tool, from C to Rust. Both versions passed nearly all of HAProxy's regression tests, but Opus 5.5 completed the task in 9.5 hours compared to 12 hours for Fable 5.1, and cost 51% less.

Eric Paulsen, field CTO for the EMEA region at Coder, expressed confidence in Opus 5.5's broader capabilities but said the development was predictable given that AI is "eating the software delivery chain end to end" today. "Despite the efficiencies showcased in Opus 5.5, the moment an agent can reliably finish real work, the question stops being whether the model is good enough and becomes where you're letting it run," Paulsen said.

Paulsen cautioned that running Claude Code sessions on personal devices poses infrastructure risks. "Kicking off a Claude Code session on a laptop that can be compromised, stolen, or simply run out of compute is not an engineering environment; capable agents need dedicated, governed infrastructure with the same guardrails, secrets handling, and observability a developer would demand of any other production workload. As we've seen here with Anthropic's own work, firms need to underline testing and safeguard procedures for launches of this kind," he added.

Anthropic confirmed that external testing was conducted by Frontier Design and METR before release. When safeguards for Opus 5.5 are triggered, requests "fall back to another model transparently," meaning developers may not know which model processed their request.

For cybersecurity tasks, users can identify and fix bugs, but most cybersecurity work routes to Opus 4.8. Requests flagged by biology and frontier LLM development classifiers redirect to Opus 5. Organizations can apply to Anthropic's Life Sciences Verification Program to use Opus 5.5 for biology research, and the company plans to expand its Cyber Verification Program in the coming weeks.

The terminal as model workspace

Sergey Ermakovich, co-founder of HasData, observed a fundamental shift in how developers interact with AI models. "The biggest change — and it's a trend that will have driven Anthropic's design and development aspirations for Claude Opus 5.5 — is that the terminal is no longer just a place where a developer pastes generated code," Ermakovich said. "The terminal now becomes part of the model's workspace."

Because the model can execute commands, examine failures, modify files and confirm results, Ermakovich suggested it can "close the loop" rather than returning incomplete work to a developer. "That makes small complete tasks much more valuable. Fix one failing test, update one dependency, migrate one endpoint, verify it, then move to the next task," he added.

John Ruelas, staff software engineer at Ramp, said verbose and difficult-to-follow output had frustrated him with frontier models, but "Claude Opus 5.5 fixes it" for his work. He noted that the model writes like a capable colleague and adheres to his company's writing standards. A design specification emerged usable with minimal revision, and when it rewrote one of his prompts, he preferred its version to his original. After the model optimized his team's test suite, he could follow its logic clearly and "shipped the change with confidence."

Beyond autocomplete

Maxime Vermeir, vice president of AI strategy at Abbyy, characterized Opus 5.5's lifecycle-spanning features as marking the end of the autocomplete era for code automation. "Anthropic's elevation here reflects the fact that it used to be thought of as marvelous if AI could complete a developer's next line of code, but today the expectation is that you hand it a whole Jira ticket and that it gets done," Vermeir said. "But despite the power on show with Claude Opus 5.5, the question remains as to how much these newer models will actually understand what 'done' means, as often it seems they have a desire to keep burning tokens by offering you yet another thing it didn't quite do right."

Beyond task completion, Opus 5.5 "communicates more naturally" than earlier versions, with early testers noting clearer and more readable output that makes it a stronger partner across extended work sessions.

Pricing for Opus 5.5 sits below Opus 5 at $4 per million input tokens and $20 per million output tokens, representing a 20% reduction from Opus 5. Anthropic also cut cache read prices by 60% for token-billed usage. The model requires fewer tokens to produce higher-quality results and generates output more than 30% faster than Opus 5.