Anthropic on Tuesday unveiled Claude Sonnet 5, describing it as the most capable model yet in its Sonnet family. According to the company, the new model can carry out agentic tasks autonomously - including planning, using tools and writing code - at levels of performance that previously required more expensive offerings.
Anthropic said Sonnet 5 approaches the performance of its Opus 4.8 model while remaining positioned at a lower price point. The company reported that Sonnet 5 shows measurable gains over Sonnet 4.6 across several dimensions, notably reasoning, tool use, coding, and knowledge-worker tasks.
Safety assessments conducted by Anthropic indicated Sonnet 5 produces a lower overall incidence of undesirable behaviors when compared with Sonnet 4.6. The company also noted that Sonnet 5 demonstrates reduced cybersecurity capabilities relative to current Opus models.
Deployment of the model began Tuesday: Sonnet 5 is the default model for Free and Pro subscribers. Team and Enterprise customers can access Sonnet 5 through Claude Code and via the Claude Platform.
Anthropic introduced introductory usage pricing for Sonnet 5 that will remain in effect through August 31, 2026: $2 per million input tokens and $10 per million output tokens. The company said pricing will rise after that date to $3 per million input tokens and $15 per million output tokens.
On comparative evaluations of agentic search and computer use, Anthropic reported Sonnet 5 outperformed Sonnet 4.6 across various effort levels. The company also emphasized that while Sonnet 5 improved on those tasks relative to Sonnet 4.6, Opus 4.8 still delivers higher accuracy for the same tasks but at a higher price point.
In targeted safety testing, Anthropic found Sonnet 5 more likely than Sonnet 4.6 to refuse malicious requests and to resist prompt injection attacks. The company reported lower rates of hallucination and sycophancy in Sonnet 5 compared with its predecessor.
Anthropic's cybersecurity evaluations showed Sonnet 5 was unable to produce working exploits for software vulnerabilities, although it registered slightly higher partial success rates than Sonnet 4.6 in those tests. The company said it launched Sonnet 5 with cyber safeguards enabled by default.
Availability
- Default model for Free and Pro subscription tiers.
- Accessible to Team and Enterprise customers via Claude Code and the Claude Platform.
Pricing
- Introductory rates through August 31, 2026: $2 per million input tokens; $10 per million output tokens.
- Post-introductory rates: $3 per million input tokens; $15 per million output tokens.
This announcement outlines Anthropic's stated improvements in capability, safety and deployment choices for Sonnet 5 while noting where limitations remain compared with the company's higher-end Opus models.