Claude Opus is 📝Anthropic's high-capability model tier within the 📝Claude family — its most capable line from the Claude 3 launch in March 2024 until the 📝Claude Mythos tier arrived above it in 2026. The current Opus model, 📝Claude Opus 5.5, is Anthropic's recommended starting point for most workloads, built for long-running agentic coding and knowledge work.
Model Lineage
The Opus tier was established alongside the Sonnet/Haiku naming convention in the Claude 3 family. Key releases:
- Claude 3 Opus (March 2024) — original flagship; $15/$75 per million tokens (input/output)
- Claude Opus 4 (May 2025) — introduced alongside Claude Sonnet 4; the first Anthropic model deployed under AI Safety Level 3 protections
- Claude Opus 4.1 (August 2025) — incremental update with enhanced agentic precision
- Claude Opus 4.5 (November 2025) — major coding leap; 80.9% SWE-bench Verified; cut Opus pricing to $5/$25
- Claude Opus 4.6 (February 2026) — longest task-completion horizon of any model at the time (METR: 50%-horizon of 14h 30m)
- Claude Opus 4.7 (April 16, 2026) — introduced the current tokenizer, the
xhigheffort level, and 3x vision resolution - Claude Opus 4.8 (May 28, 2026) — final Claude 4 Opus; now the fallback model for cybersecurity requests that Mythos-class safeguards decline
- Claude Opus 5 (July 24, 2026) — first Opus of the Claude 5 generation; $5/$25
- Claude Opus 5.5 (September 22, 2026) — current flagship Opus; $4/$20, always-on thinking, Mythos-class safeguards
Claude Opus 5.5 (Current)
Claude Opus 5.5 is the first model in the Claude 5.5 family. Anthropic reports that it performs at the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5, and it is the first Opus model to launch with safeguards similar to Fable 5.1's on cybersecurity, biology, and distillation. Its thinking is always on, its default effort is medium, and it carries a 1M-token context window with 128K max output. See 📝How to Prompt Claude Opus 5.5 and 📝Claude Opus 5 vs Claude Opus 5.5.
Claude Opus 4.7
At its April 2026 release, Opus 4.7 was positioned as the production-safe frontier model, explicitly less capable than Claude Mythos Preview but incorporating Mythos-era safeguards for real-world deployment.
Core improvements over 4.6:
- Vision resolution increased 3x — up to 2,576px / 3.75 megapixels
- Self-verification: catches logical faults during the planning phase
- New
xhigheffort level, betweenhighandmax, for granular reasoning depth control - Task budgets (public beta): hard ceiling on token spend for agentic runs
- CursorBench coding: 70% vs 58% for Opus 4.6
- GDPVal-AA knowledge work: Elo 1753 (GPT-5.4: 1674, Gemini 3.1 Pro: 1314)
- Stricter instruction-following; improved prompt injection resistance
- New
/ultrareviewcommand in 📝Claude Code — simulates a senior reviewer, flags design flaws and logic gaps
Pricing: $5/$25 per million tokens (input/output) — unchanged from 4.6. Up to 90% savings with prompt caching, 50% with batch. 300k output tokens available via batch API with output-300k-2026-03-24 beta header.
Access: Still available as a legacy model through the API, Amazon Bedrock, 📝Google Cloud Vertex AI, and Microsoft Foundry. Model string: claude-opus-4-7. Retirement not sooner than April 16, 2027.
Context: 1M-token context window; up to 128K tokens output (synchronous).
Positioning
Opus is Anthropic's recommended tier for demanding production work: agentic coding, sophisticated AI agents, complex document creation, and long-horizon autonomous workflows. The price-per-capability ratio has compressed dramatically — Claude 3 Opus cost $15/$75; Opus 5.5 delivers far more capability at $4/$20. Since June 2026, the most demanding reasoning has moved up to the Mythos tier, while Opus 5.5 closes most of that gap at 40% of Fable 5.1's price.
Cybersecurity & Safety
Opus 4.7 served as the testbed for Anthropic's automated safeguards targeting high-risk cybersecurity use, with cyber capabilities deliberately scoped below Mythos Preview. Anthropic launched the Cyber Verification Program alongside 4.7, allowing legitimate security professionals (pen testers, red-teamers, vuln researchers) to access elevated capabilities through credential verification. Opus 5.5 extends that posture with classifiers for cybersecurity, biology, and reasoning extraction.