Claude Sonnet 5.5
The second model of the Claude 5.5 wave and Anthropic's fastest Sonnet yet — a lower-cost complement to Opus 5.5 for well-scoped everyday tasks, bug fixing, and polished documents, slides, and spreadsheets
Claude Sonnet 5.5
Anthropic • September 2026
Training Data
Up to June 2026
Claude Sonnet 5.5
September 2026
Parameters
Not disclosed
Training Method
Advanced Constitutional AI with adaptive thinking (high effort by default), cyber safeguards that fall back to Sonnet 5 on high-risk requests, and reasoning-extraction classifiers
Context Window
1,000,000 tokens
Knowledge Cutoff
June 2026
Key Features
Near-Opus 5.5 Performance • 30%+ Faster Output • 1M Context / 128K Output
Capabilities
Agentic Coding: Outstanding (70.6% Terminal-Bench 4.0)
Computer Use: Outstanding (80.1% OSWorld 2.1)
Cost Efficiency: Excellent
What's New in This Version
Edges past Opus 5.5 on Terminal-Bench 4.0 (70.6% vs 66.4%) and lifts OSWorld 2.1 from 57.0% to 80.1% and CursorBench 4.0 from 34.1% to 55.5% over Sonnet 5, at the same $2/$10 per 1M tokens and up to 30% less per task; the first Sonnet with frontier-style cyber safeguards, and the first to beat Pokémon Red from screenshots alone
The second model of the Claude 5.5 wave and Anthropic's fastest Sonnet yet — a lower-cost complement to Opus 5.5 for well-scoped everyday tasks, bug fixing, and polished documents, slides, and spreadsheets
What's New in This Version
Edges past Opus 5.5 on Terminal-Bench 4.0 (70.6% vs 66.4%) and lifts OSWorld 2.1 from 57.0% to 80.1% and CursorBench 4.0 from 34.1% to 55.5% over Sonnet 5, at the same $2/$10 per 1M tokens and up to 30% less per task; the first Sonnet with frontier-style cyber safeguards, and the first to beat Pokémon Red from screenshots alone
Technical Specifications
Key Features
Capabilities
Other Anthropic Models
Explore more models from Anthropic
Claude Haiku 5.5
The final model of the Claude 5.5 wave and Anthropic's cheapest, fastest, most capable small model, built for high-volume, latency-sensitive work such as classification, extraction, routing, and computer use
Claude Opus 5.5
Anthropic's recommended default for long-running agentic coding and knowledge work — Fable 5.1-level performance on most tasks at Opus pricing, and the first release of the 5.5 wave
Claude Fable 5.1
Anthropic's most intelligent generally available model, built for demanding reasoning and long-horizon agentic work that fixes root causes rather than symptoms
