Claude Sonnet 5
Claude Sonnet 5 is Anthropic’s most agentic Sonnet model. It plans, uses tools, and completes multi-step coding and knowledge tasks, narrowing the gap to Opus-class performance while improving safety and cost efficiency across browsing, computer-use, and software workflows.
Overview
Define the objective, select allowed tools, and set the effort level. Sonnet 5 decomposes the task, browses for context, executes computer actions or terminal steps, writes and tests code, self-reviews outputs, and returns artifacts or structured results, while tracking token use to manage budget and latency.
Capabilities and performance
Best for teams building production agents, IDE copilots, and workflow automation across software engineering, data operations, and knowledge work. It serves product teams needing predictable cost-performance, enterprises standardizing on safer defaults, and researchers evaluating agent reliability. If you require maximal raw capability or reduced guardrails for security research, an Opus model may be preferable.
- Plans tasks, chooses tools, and executes multi-step workflows with self-review loops.
- Operates browsers and terminals programmatically to gather context and perform actions.
- Generates, edits, and tests code changes across files, frameworks, and repositories.
- Balances cost and quality using selectable effort levels and token budgets.
- Refuses unsafe requests and enforces real-time cybersecurity safeguards by default.
Key capabilities
Who it's for
Use the model identifier claude-sonnet-5 via the Claude API, Claude Code, or the Claude Platform. Select effort levels, set tool permissions, and define structured inputs and outputs for automation. Increased rate limits support higher-token work at high effort. Note the updated tokenizer may map the same text to more tokens; monitor budgets accordingly. For enterprise rollouts, enable safeguards, log tool calls and decisions, and start with representative BrowseComp-style and OSWorld-Verified-style tasks to calibrate effort, latency, and cost before broad deployment.
In practice, Sonnet 5 finishes what it starts—planning, checking, and shipping durable results at a price most teams can run every day.
Getting started
Sonnet 5 pairs near-Opus capability with Sonnet economics, reliable tool use, and safer defaults. It finishes complex sequences, respects constraints, and surfaces verifiable outputs. Adjustable effort levels and clear token tracking let teams tune latency and spend. Choose it for everyday agents, browsing, and coding work; prefer Opus when maximum raw capability or looser cybersecurity guardrails are explicitly required.
Open the tool and review its core product experience.
Create your account or access your existing workspace.
Use your own task to judge speed, quality, and fit.
Check similar AI tools before making a final decision.
Claude Sonnet 5
Anthropic is an AI research and product company focused on helpful, honest, and harmless systems. The Claude family spans Mythos, Fable, Opus, Sonnet, and Haiku, supporting agents, coding, and enterprise workflows. The team emphasizes safety evaluations, deployment guardrails, and practical cost-performance for real-world use.
Claude Sonnet 5 is a versatile, agentic large language model tuned for end-to-end execution. It plans steps, invokes tools such as browsers and terminals, and autonomously carries tasks through review and verification. Compared with Sonnet 4.6, it delivers stronger reasoning, more reliable tool use, and better cost-performance; on common agentic evaluations (BrowseComp, OSWorld-Verified) it reaches capability levels that previously required larger, pricier models. Sonnet 5 is available across Claude Free, Pro, Max, Team, and Enterprise plans, in Claude Code, and via the Claude API. Introductory pricing runs through August 31, 2026 ($2/MTok input, $10/MTok output), then standard pricing applies ($3/$15). Safety evaluations show lower rates of undesirable behaviors versus Sonnet 4.6 and substantially weaker offensive cybersecurity ability than Opus-class models; real-time cyber safeguards ship enabled by default. A tokenizer update improves modeling fidelity while mapping some texts to more tokens (roughly 1.0–1.35x), with launch pricing designed to keep transitions cost-neutral. Choose effort levels to trade speed, quality, and budget per task.
www.anthropic.com/news/claude-sonnet-5
Comments (0)
No Comments Found