Claude Opus 5.5

Overview

Anthropic’s September 22, 2026 flagship Opus-tier model — first release in the Claude 5.5 family. API ID claude-opus-5-5. Positioned for long-running agentic coding and knowledge work at Fable 5.1-parity with 40% lower compute cost vs claude-opus-5.

Pricing

Tier	Opus 5.5	Opus 5
Input	$4/MTok	$5/MTok
Output	$20/MTok	$25/MTok
Cache read	$0.20/MTok	$0.50/MTok

Breaking API Changes (from Opus 5)

1. Thinking cannot be disabled — use effort parameter
2. Forced tool use returns error
3. Thinking blocks tied to model and conversation
4. computer_20251124 computer use tool deprecated (Claude API + Google Cloud)

Production integrations on Opus 5 require migration review before switching model ID.

Benchmarks

• Artificial Analysis Intelligence Index: 58 at max effort (vendor benchmark, highest measured)
• Parity with gpt-6-astra on Terminal-Bench 4.0 and AutomationBench-AA per Artificial Analysis
• Anthropic claims SOTA coding/knowledge work; attribute to vendor evals unless independently reproduced

Safeguards

Biology and cybersecurity capabilities comparable to claude-fable-5 — same export/safety restrictions. Pre-release evaluation by metr and Frontier Design.

Related

• anthropic
• claude-opus-5
• claude-fable-5
• frontier-models
• coding-agents
• ai-price-wars
• pacing-the-frontier

Sources

• 2026-09-22-anthropic-claude-opus-5-5-release
• 2026-09-22-claude-opus-5-5-anthropic-official
• 2026-09-22-claude-opus-5-5-artificial-analysis
• 2026-09-22-claude-opus-5-5-platform-docs