This page may contain stale information. Last updated: 2026-07-01
Overview
Claude Sonnet 5 is anthropic’s mid-tier frontier model released June 30, 2026, succeeding claude-sonnet-4-6. Positioned as the most agentic Sonnet model — planning, browser/terminal tool use, and autonomous coding at near-claude-opus-4-8 performance at lower cost.
Specifications
- API identifier:
claude-sonnet-5 - Default availability: Free and Pro plans; Max, Team, Enterprise; claude-code
- Tokenizer: New tokenizer — ~30% more tokens for same text vs Sonnet 4.6 (2026-06-30-anthropic-claude-sonnet-5-datacamp)
- Intro pricing (through Aug 31, 2026): 10/M output tokens
- Standard pricing: 15/M output tokens
Benchmarks (Anthropic-reported)
| Benchmark | Sonnet 5 | Opus 4.8 | Sonnet 4.6 |
|---|---|---|---|
| Agentic coding | 63.2% | 69.2% | 58.1% |
| Knowledge work | ~matches Opus | baseline | lower |
Anthropic corrected initial BrowseComp cost-performance chart methodology error June 30 ( 2026-06-30-anthropic-claude-sonnet-5-launch).
Safety Profile
- Lower misalignment/desirable-behavior rate than Sonnet 4.6
- Much lower offensive cybersecurity capability than Opus/Mythos
- Better prompt-injection resistance and refusal consistency (Lovable cited)
Recent Developments
- 2026-06-30: Global launch; default for consumer tiers (2026-06-30-anthropic-claude-sonnet-5-launch)
- 2026-06-30: TechCrunch frames agentic capability as new price-tier baseline vs openai GPT-5.6 Sol and gemini 3.5 Flash (2026-06-30-anthropic-claude-sonnet-5-techcrunch)