Claude Opus 5.5: Anthropic's New Flagship Is Now on VEGA AI
Anthropic released Claude Opus 5.5, the first model in the Claude 5.5 family, and it is already live on VEGA AI. On most work it performs at the level of Claude Fable 5.1 while being noticeably cheaper to run.
What changed since Opus 5
The coding numbers from early testers are the most striking part. One of them ran a 680,000-line code migration in under a day, a job that would have kept an engineering team busy for weeks. Another audited and fixed a 200,000-line codebase in under three hours; Opus 5 needed over 20 hours for the same audit and burned 2.5x the tokens.
Efficiency is the other half of the story. Opus 5.5 needs less compute to serve and spends fewer tokens per task, which comes out to roughly 40% lower cost on typical workloads. Output also streams more than 30% faster.
Anthropic also reworked how the model writes. Responses put the important part first, drop jargon, and follow whatever style rules you set. The company says this was the most common complaint about Opus 5, and early testers report the difference is easy to feel in long sessions.
On the safety side, Opus 5.5 posted the best scores Anthropic has recorded on its automated behavioral audit. It attempted to circumvent containment boundaries about 85% less often than Opus 5, and every attempt it did make was low severity and self-reported. It is also more resistant to prompt injection than Opus 5.
Benchmarks
Opus 5.5 leads Anthropic's evaluations in agentic coding, computer use, and knowledge work:
| Benchmark | Opus 5.5 | Fable 5.1 | Opus 5 | GPT-6 Astra | GPT-5.6 Sol |
|---|---|---|---|---|---|
| Terminal-Bench 4.0 (agentic coding) | 66.4% | 55.8% | 52.3% | 57.9% | 37.3% |
| FrontierCode v1.1 (Main) | 54.4% | 50.3% | 48.0% | 53.3% | 47.5% |
| CursorBench 4.0 | 57.8% | 51.8% | 46.6% | — | 41.7% |
| GDPval-AA v2.1 (knowledge work) | 1846 | 1735 | 1708 | 1542 | 1588 |
| AutomationBench (business workflows) | 40.0% | 31.4% | 26.9% | 41.4% | 28.8% |
| Humanity's Last Exam (with tools) | 67.7% | 65.6% | 63.6% | 57.2% | — |
| OSWorld 2.0 (computer use) | 81.8% | 80.7% | 74.0% | — | — |
| Chartography (chart recognition) | 89.0% | 88.4% | 83.4% | — | — |
Anthropic admits that at this level the margins mean less than they used to: in its own day-to-day use, the gap between Opus 5.5 and Fable 5.1 feels narrower than the table suggests. The clearer win is cost. At default effort, Opus 5.5 beats GPT-6 Astra on FrontierCode for about a fifth of the cost per task.
Security for autonomous agents
For teams that let agents run unattended for hours, Opus 5.5 ships with a classifier that screens every action before it executes, an open-source sandbox that security teams can audit, and code review that flags vulnerabilities before they merge. On Gray Swan's prompt injection benchmark it ties Fable 5.1 for the lowest success rate of any model tested.
Research and analysis
In an internal test, Anthropic had Opus 5.5, Fable 5.1, and Opus 5 each write a company's quarterly report using only hard-to-find sources on a copy of the web, with an automated grader checking every figure. Sixteen of 18 Opus 5.5 reports passed; the other two models never did. In a separate merger-analysis exercise, Opus 5.5 finished in 63 minutes versus 93 for Opus 5, and produced a more complete financial model along with a cleaner presentation.
Safeguards
Because Opus 5.5 is comparable to Claude Mythos 5.1 in biology and cybersecurity, it launches with the same class of safeguards as Fable 5.1: preserved thinking to resist distillation attacks, watermarking for EU AI Act compliance, and zero data retention. Vetted organizations can apply to Anthropic's Life Sciences Verification Program, and the Cyber Verification Program opens to Opus 5.5 in the coming weeks.
Try it on VEGA AI
Claude Opus 5.5 is available in the VEGA AI model picker right now. Claude Sonnet 5.5 and Claude Haiku 5.5 will follow in the coming weeks.