Claude Sonnet 5 is designed to be Anthropic's most agentic Sonnet model to date. It can formulate plans, operate tools such as browsers and terminals, and function autonomously at a level that, only months earlier, required larger and costlier models.
For many developers, the agentic AI era kicked off with Sonnet-class models: Claude Sonnet 3.5, 3.6, and 3.7 were the first to demonstrate strong coding and tool-use abilities. More recently, however, the biggest leaps in agentic capability have come from Anthropic's Opus-class models.
Sonnet 5 closes much of that gap: its performance approaches that of Opus 4.8 but at a lower price. It represents a major step up from its predecessor, Sonnet 4.6, across key dimensions of agentic performance including reasoning, tool use, coding, and knowledge work.
Anthropic's pre-deployment safety assessments found that Sonnet 5 exhibits a lower overall rate of undesirable behaviors than Sonnet 4.6 and is generally safer to deploy in agentic contexts. Evaluations also indicate that it has substantially weaker cybersecurity task capabilities compared to the current Opus models.
As of launch day, Claude Sonnet 5 is available across all plans: it serves as the default model for Free and Pro plans and is accessible to Max, Team, and Enterprise users. It is also available in Claude Code and on the Claude Platform with introductory pricing of $2 per million input tokens and $10 per million output tokens through August 31, 2026, after which standard pricing of $3 per million input tokens and $15 per million output tokens takes effect. Developers can access it via the Claude API using the identifier claude-sonnet-5.
Working with Claude Sonnet 5
Performance comparisons of Sonnet 5 with Sonnet 4.6 and Opus 4.8 at varying effort levels on the agentic search benchmark BrowseComp and the computer use benchmark OSWorld-Verified show that Sonnet 5 strictly outperforms Sonnet 4.6. While Opus 4.8 remains the top choice for maximum accuracy on these tasks, Sonnet 5 gives developers lower-cost options at much higher quality than was previously available. Between Sonnet 5 and Opus 4.8, users can tune the effort level to find the ideal cost-performance tradeoff.
Feedback from early access partners has been consistent: Sonnet 5 is far more agentic than its predecessors. Testers reported that it completes complex tasks where earlier Sonnet models would stall, that it self-checks its output without being explicitly instructed to do so, and that it delivers all of this at an attractive price point.
Early partner testimonials highlighted its strengths across varied use cases:
- Strong execution for multi-step software engineering workflows, handling sustained coding, tool use, and debugging across messy technical contexts.
- Reliable end-to-end completion of compound automation tasks (e.g., updating Salesforce tiers and sending launch announcements) that previously stalled midway.
- Achieving the same output quality with fewer steps, while consistently and cleanly refusing unsafe requests.
- Carrying challenging real pull requests through to tested, verified results autonomously.
- Proactively writing reproducing tests, implementing fixes, and verifying them-all in a single pass.
- Staying on plan, following conventions, and shipping clean multi-step changes at efficient cost.
- Excelling on brownfield code-race conditions, hidden tests-by tracing failures to root causes and shipping durable fixes rather than superficial patches.
- Sitting on the Pareto frontier for legal research and analysis tasks at a compelling price-to-performance ratio.
- Reasoning in tighter steps and getting users to answers noticeably faster in data exploration workflows.
- Consistently taking the correct action quickly in insurance workflow automation.
Safety evaluations
Anthropic's pre-deployment safety evaluations found Sonnet 5 to be an overall improvement over Sonnet 4.6. On agentic safety, the model is better at declining malicious requests and resisting hijack attempts in prompt injection attacks. It exhibits lower rates of hallucination and sycophancy than Sonnet 4.6. On Anthropic's automated behavioral audit-which tests a broad range of misaligned behaviors such as cooperation with misuse and deception-Sonnet 5 scored lower (i.e., safer) overall. However, it did display somewhat higher rates of misaligned behavior on this assessment compared to the more capable Opus 4.8 and Claude Mythos Preview.
Sonnet 5 was not deliberately trained on cybersecurity tasks. It can handle some routine, non-harmful cyber tasks, but on evaluations testing potentially dangerous cyber skills-such as developing software exploits-it performs substantially worse than models like Opus 4.8 and Mythos 5. On a Firefox vulnerability exploit evaluation (developed in collaboration with Mozilla, with all vulnerabilities patched in Firefox 148), Sonnet 5 was never able to produce a full working exploit, though it showed a slightly higher rate of partial success compared to Sonnet 4.6. This change is likely attributable to general intelligence improvements rather than specific training.
Because Sonnet 5 is somewhat stronger than its predecessor on these tasks, Anthropic has launched it with cyber safeguards enabled by default. These safeguards-which detect and block dangerous cyber usage in real time-are the same as those present in Claude Opus 4.7 and 4.8. Because the overall cybersecurity risk from Sonnet 5 was judged to be low, the safeguards are less strict than those launched with Fable 5, which block a wider range of cybersecurity tasks.
Anthropic's full assessment of Sonnet 5 across many safety and capability evaluations is reported in the Claude Sonnet 5 System Card.
Availability and pricing
Claude Sonnet 5 is available everywhere at an introductory price of $2 per million input tokens and $10 per million output tokens through August 31, 2026, moving to standard pricing of $3 per million input tokens and $15 per million output tokens afterward. Rate limits across Chat, Cowork, Claude Code, and the Claude Platform have been increased to accommodate the higher token usage of higher effort levels; users can select whichever level suits their project.
Additional notes:
- Sonnet 5 is part of Anthropic's Cyber Verification Program, available on the native Claude Platform, the Claude Platform on AWS, and Claude in Microsoft Foundry (hosted on Azure and Anthropic), and coming soon on Claude in Google Vertex. Organizations already enrolled automatically have the same access on Sonnet 5 without needing to reapply. Claude Opus 4.8 is recommended for cybersecurity work requiring reduced guardrails.
- Sonnet 5 is an upgrade to Sonnet 4.6 but uses an updated tokenizer that changes how the model processes text to improve performance (similar to the tokenizer change introduced with Claude Opus 4.7). The tradeoff is that the same input can map to roughly 1.0–1.35× more tokens depending on content type. The introductory pricing is set so the transition is roughly cost-neutral.
- On April 26, 2026, Anthropic raised Sonnet and Haiku rate limits at every usage tier and simplified to three tiers (Start, Build, and Scale) on the native Claude Platform.