Claude Sonnet 5 Release Notes

Claude Sonnet 5 enhances agentic capabilities and tool use

Claude Sonnet 5 is designed as the most agentic model in the Sonnet family, enabling it to create plans, utilize browsers and terminals, and operate autonomously. It significantly narrows the performance gap between the Sonnet-class and Opus-class models, offering capabilities close to Opus 4.8 while maintaining a lower cost structure.

Key Improvements over Sonnet 4.6

Sonnet 5 provides substantial improvements in reasoning, coding, knowledge work, and tool use. Early access partners report that the model is more capable of completing complex, multi-step tasks without stalling, and it frequently performs self-correction and output verification unprompted.

Specific real-world applications highlighted by partners include:

• Software Engineering: Handling sustained coding, debugging, and tracing failures to root causes in "brownfield" code (legacy codebases).
• Automation: Completing end-to-end workflows, such as updating Salesforce account tiers and sending launch announcements.
• Legal Research: Improving legal research and analysis for plaintiff-law tasks.
• Data Analysis: Reducing time-to-insight by reasoning in tighter steps when exploring live data.
• Insurance Workflows: Executing submission intake and loss runs on existing operational systems.

Performance Benchmarks

In evaluations using BrowseComp (agentic search) and OSWorld-Verified (computer use), Sonnet 5 is a strict improvement over Sonnet 4.6. While Opus 4.8 remains the superior choice for maximum accuracy, Sonnet 5 allows developers to balance cost and performance by adjusting the "effort" level.

Safety and Cybersecurity Guardrails

Sonnet 5 exhibits a lower rate of undesirable behaviors and hallucinations compared to Sonnet 4.6, making it safer for agentic contexts. It is more resistant to prompt injection attacks and better at refusing malicious requests.

Cybersecurity Limitations

Anthropic did not deliberately train Sonnet 5 on cybersecurity tasks. Consequently, it performs substantially worse than Opus 4.8 and Mythos 5 on dangerous cyber skills, such as developing software exploits. In tests involving Firefox 147 vulnerabilities, Sonnet 5 was unable to develop a full working exploit, though it showed a slightly higher partial success rate than Sonnet 4.6 due to general intelligence gains.

Because of this slight increase in capability, Sonnet 5 is launched with real-time cyber safeguards enabled by default to detect and block dangerous usage.

Availability and Pricing

Claude Sonnet 5 is available to all users across Free, Pro, Max, Team, and Enterprise plans. It is also integrated into Claude Code and the Claude Platform.

API Pricing

To account for a tokenizer change that increases token counts by approximately 1.0–1.35×, Anthropic has introduced introductory pricing to keep the transition cost-neutral:

Period	Input Tokens (per million)	Output Tokens (per million)
Through August 31, 2026	$2	$10
After August 31, 2026	$3	$15

Developers can access the model via the API using the identifier claude-sonnet-5.