Claude Sonnet 5

2026-06-30 · Publisher : Anthropic · Domain : Artificial Intelligence

Anthropic launches Claude Sonnet 5, its most agentic Sonnet: planning, use of browsers and terminals, autonomous execution at a level that recently required larger and more expensive models.

What changes

  • Performance close to Opus 4.8 at lower cost, with more reasoning, tool use, coding and knowledge work.
  • Reduced cyber capabilities compared with the Opus models; undesirable behaviors are rarer.
  • New tokenizer.
  • Default model of the Free and Pro plans.

### Specifications - Context window: 1M tokens. Maximum output: 128,000 tokens (models page). - Knowledge cutoff: not disclosed for this launch. - Availability: Claude API and other platforms. API identifier: `claude-sonnet-5`. ### Pricing | Item | Price (USD per million tokens) | | --- | --- | | Input | 2 | | Output | 10 | | Cache write 5 min / 1 h | 2.50 / 4 | | Cache read | 0.20 | | Batch (input / output) | 1 / 5 | At announcement: introductory pricing until August 31, 2026 before a planned move to $3 / $15. That move will not take place: $2 / $10 has become the standard price (pricing page). ### Benchmarks Scores published by the vendor: the blog post mainly gives Sonnet 4.6 values as a reference (Humanity's Last Exam without tools 34.6%, with tools 46.8%; OSWorld-Verified 78.5%). The scores of Sonnet 5 itself are not recorded. On Terminal-Bench 4.0, the Sonnet 5.5 page gives 10.3% for Sonnet 5. ### Key takeaways Sonnet 5 lowers the price of the mid-range tier from $3 / $15 to $2 / $10 while targeting agentic uses.

Links

Tags : LLM, Anthropic, Claude, Sonnet, Agents, Contexte 1M