Claude Opus 5.5 Launches with 40% Lower Cost and 30% Faster Output

Anthropic has launched Claude Opus 5.5, its latest large language model, optimized for programming, agentic workflows, and demanding professional tasks — all at a lower operational cost than its predecessor.

In an official statement, the company said the new model approaches the performance of Claude Fable 5.1 across most benchmarks while cutting runtime expenses by approximately 40% compared to Claude Opus 5 under typical workloads.

Anthropic described Opus 5.5 as the strongest Opus-tier model released to date and the first in the newly introduced Claude 5.5 family.

The company set API pricing at $4 per million input tokens and $20 per million output tokens — down from $5 and $25, respectively, for Opus 5. Cache read costs fell to $0.20 per million tokens, representing a 60% reduction.

Opus 5.5 generates outputs more than 30% faster than Opus 5. Its Fast Mode, available in both Claude Code and the Claude platform, operates at 2.5 times the speed of the prior version.

Complex task execution

Opus 5.5 is engineered specifically for long-horizon, intricate tasks — particularly agentic coding. Anthropic stated the model can carry out large-scale codebase migrations, identify bugs, conduct code reviews, and manage multi-tool operations over extended durations with minimal human intervention.

According to internal testing, Opus 5.5 scored 66.4% on Terminal-Bench 4.0, a benchmark for agentic programming, surpassing Opus 5’s 52.3% and GPT-6 Astra’s 57.9%. It achieved 54.4% on FrontierCode v1.1 and 1,846 points on GDPval-AA v2.1, a test designed to assess real-world cognitive task performance.

Rigorous safety protocols

Although Anthropic emphasized performance gains, safety remained a central pillar of the Opus 5.5 release.

The company confirmed Opus 5.5 is the first Opus-class model launched with safety assurances comparable to those applied to Claude Fable 5.1 in cybersecurity, biological weapons, and anti-model-stealing domains. Prior to launch, the model underwent external evaluations conducted by METR and Frontier Design.

Anthropic explained that Opus 5.5 incorporates pre-execution action verification, an open-source isolated runtime environment accessible for security team review, and a dedicated code review system designed to detect vulnerabilities before integration.

The model demonstrated stronger resistance to prompt injection attacks than Opus 5 across tests covering coding, tool use, web browsing, and computer control.

It also attained the highest score to date within Anthropic’s proprietary automated behavioral alignment evaluations — simulations involving thousands of scenarios to measure adherence to defined operational boundaries.

Claude Opus 5.5 is now available to users on Pro, Max, Team, and Enterprise plans via the Claude application, to developers through the Claude API and Claude Code, and via Amazon Web Services, Google Cloud, and Microsoft Foundry.

Anthropic plans to release Claude Sonnet 5.5 and Claude Haiku 5.5 later this month as part of the same 5.5 family rollout.