Anthropic Launches Claude Opus 5.5 to Lead AI Benchmarks

Anthropic has launched Claude Opus 5.5, a powerful model that outperforms its predecessor Claude Opus 5, Claude Fable 5.1, and OpenAI's GPT-6 Astra. Released alongside OpenAI's GPT-6 Sol and GPT-6 Luna, the model processes up to 1 million input tokens of text and images, and outputs up to 128,000 tokens, or 300,000 via the Batch API. It features always-on reasoning across five levels—low, medium, high, xhigh, and max—defaulting to high. It also includes statistical watermarking and a fast mode offering 2.5 times the speed at double the cost. For sensitive cybersecurity and biology queries, it falls back to Claude Opus 4.8.

For practitioners, the model introduces zero data retention, unlike Fable 5.1's 30-day policy. API pricing is $4 per million input tokens, $0.25 per million cached tokens, and $20 per million output tokens. Cache reads and writes cost $0.20 and $5 per million tokens, while batch processing is $2 for inputs and $10 for outputs per million tokens. Although token prices are lower, Artificial Analysis notes high token usage results in a task cost of $5.98 at max reasoning, compared to $7.63 for Fable 5.1 and $3.26 for GPT-6 Astra.

Claude Opus 5.5 dominates benchmarks, scoring 58 on the Artificial Analysis Intelligence Index v4.3, five points ahead of GPT-6 Astra and Fable 5.1. It leads in Humanity's Last Exam at 61.4 percent, SciCode at 66.9 percent, and ties on Terminal-Bench 4.0 at 59.6 percent. On the Vals AI Index, it ranks first with 69.69 percent, leading in ProofBench v1.1 with a perfect score, MedScribe, and the RSI Index. It also passed Anthropic's internal knowledge work threshold on 16 of 18 attempts, whereas Opus 5 and Fable 5.1 failed every attempt.

Anthropic trained the model on public and private data, including ClaudeBot crawler data, synthetic data, and user inputs, using the same knowledge cutoff as Claude Fable and Mythos 5.1. The company plans to release Claude Sonnet 5.5 and Claude Haiku 5.5 in the coming weeks, marking the first Haiku update since version 4.5 in October 2025. This rollout provides developers with an articulate, instruction-compliant tool for complex data analysis and report generation.