Claude Sonnet 4.5 by Anthropic is a hybrid reasoning model built for agentic coding, computer use, and long-running tasks with extended thinking and ASL-3 safety protections.
Capabilities, design details, and architectural traits
Claude Sonnet 4.5 is a hybrid reasoning model from Anthropic, designed as the primary workhorse for agentic tasks, software engineering, and direct computer use. It operates in two modes: a fast default response mode and extended thinking mode, where it outputs a visible chain-of-thought for harder problems. It is the first Claude model deployed under AI Safety Level 3 (ASL-3) protections, with classifiers specifically targeting CBRN-related inputs and outputs.
| Trait | Detail |
|---|---|
| Dual operating modes | Toggles between standard response and extended thinking, with a visible reasoning chain for complex problems |
| 30+ hour task endurance | Documented ability to sustain focus on multi-step, long-horizon tasks across agentic loops |
| Computer use leadership | Leads on OSWorld at 61.4%, designed to navigate browsers, click buttons, fill forms, and recover from errors |
| ASL-3 deployment | Released under Anthropic's AI Safety Level 3 standard with CBRN classifiers and prompt-injection defenses for agentic use |
| Most aligned frontier model (at release) | System card includes first use of mechanistic interpretability to evaluate alignment, with documented reductions in sycophancy, deception, and power-seeking |
| Context editing and memory tool | Supports extended autonomous agent sessions via an API-level memory tool and context editing feature |
Sonnet 4.5 launched alongside the Claude Agent SDK, giving developers the same agent-building infrastructure Anthropic uses internally for Claude Code. The model is designed to be orchestrated as well as to orchestrate - Anthropic explicitly documented a pattern where Sonnet 4.5 breaks down problems and coordinates multiple Haiku 4.5 instances as sub-agents in parallel.
Independent evaluations · Artificial Analysis
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | Anthropic | Anthropic | Anthropic |
| Release Date | September 29, 2025 | July 24, 2026 | June 9, 2026 |
| Knowledge Cutoff | Jan 2025 | May 2026 | - |
| Context & Limits | |||
| Context Window | 1M | 1M | 1M |
| Pricing (per 1M tokens) | |||
| Input Pricing | $3 Best Input Pricing | $5 | $10 |
| Output Pricing | $15 Best Output Pricing | $25 | $50 |
| Modalities | |||
| Inputs | textimagefile | textimage | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 29.9 | 63.1 Best Intelligence Index | 62.1 |
| Coding Index | - | 78.0 Best Coding Index | 76.5 |
| Agentic Index | - | 59.2 Best Agentic Index | 56.6 |
Graduate-level reasoning and expert Q&A evaluation.
Extremely difficult logical reasoning and knowledge.
Logical reasoning over long context windows.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
