Anthropic announced on September 22 the release of Claude Opus 5.5: the same flagship model in the Opus family, but 40% cheaper via the API, more than 30% faster, and scoring best yet on Anthropic's own automated behavioral and alignment audit. Sonnet 5.5 and Haiku 5.5 are coming "in the coming weeks," according to the company.
Opus 5.5 replaces Opus 5 as Anthropic's most capable model, but the core change in this release isn't a dramatic new capability — it's an efficiency win: the same level of reasoning at a considerably lower cost and with faster response times. Anthropic reports that Opus 5.5's overall performance is on par with Fable 5.1, and that it's the highest-scoring model to date on the company's internal automated behavioral and alignment audit — the check Anthropic uses to evaluate whether a model tends to behave honestly, cautiously, and predictably under pressure before it ships.
The pricing comparison is the most directly actionable part for any business already building on the Claude API:
Anthropic published results across a test set oriented toward real agentic work and software development, not just academic questions: Terminal-Bench 4.0 (command-line tasks), FrontierCode v1.1 (complex code), CursorBench 4.0 (assisted code editing), GDPval-AA v2.1 (economically valuable tasks evaluated against human performance), and OSWorld 2.0 (computer/agent control of a real operating system). We won't repeat exact figures from those benchmarks here without verifying them ourselves against the primary source — but a consistent pattern across all five is that Opus 5.5 moves in the direction of "cheaper and faster without losing ground," which is the combination that actually matters for anyone running these models in production at volume.
Carlos Montiel is an enterprise AI solutions architect. He implements LLMs, Agents, RAG, and orchestrators for companies in Guatemala and Latin America. Reach out for a consultation.
Contact Carlos Montiel