Introducing Claude Sonnet 5
Key Points
- Agentic performance near Opus 4.8 at lower cost
- Intro pricing $2/$10 per million tokens through 2026-08-31
- Available as default on Free/Pro and via claude-sonnet-5 API
Summary
Claude Sonnet 5 is a next-generation agentic Sonnet model optimized for multi-step planning, tool use (browsers, terminals), and autonomous workflows. It narrows the capability gap with Opus 4.8 while offering substantially better cost efficiency than prior Sonnet releases, and ships with cyber safeguards enabled by default.
Key Points
- Performance: Marked improvement over Sonnet 4.6 in reasoning, tool use, coding, and knowledge work; in some high-effort settings it can match Opus 4.8.
- Cost & pricing: Introductory pricing through 2026-08-31 at $2 per million input tokens and $10 per million output tokens; standard pricing after that is $3/MTok input and $15/MTok output.
- Availability: Default model for Free and Pro plans; available to Max, Team, and Enterprise; accessible via the Claude API as
claude-sonnet-5, Claude Code, and the Claude Platform. - Effort tuning: Developers can adjust effort levels to trade cost for performance; Sonnet 5 shows especially good cost-performance at medium effort.
- Safety & cyber: Overall safer than Sonnet 4.6 on agentic safety metrics (better refusal, lower hallucination/sycophancy), but still shows higher misalignment than Opus 4.8; poor at developing exploits and launched with real-time cyber safeguards.
- Engineering fit: Well suited for sustained coding, debugging, brownfield fixes, multi-step automation, pull-request verification, and agentic search/computer-use agents.
- Operational: Rate limits increased across Chat, Cowork, Claude Code, and the Platform to support higher-effort usage.
Practical notes for engineers
- API: Use model name
claude-sonnet-5via the Claude API; tune effort level to balance latency, token consumption, and capability. - Safeguards: Cybersecurity-related tool usage is blocked/monitored by default; do not rely on Sonnet 5 for exploit or offensive cyber tasks.
- Audit: See the Claude Sonnet 5 System Card for detailed evals, safety results, and the corrected cost-performance methodology (changelog: chart update on 2026-06-30).