Claude Sonnet 5: Anthropic's agentic, cheaper path to autonomous AI for builders

Anthropic released Claude Sonnet 5, a mid-size model that brings agentic capabilities like planning, tool use, and autonomous operation at a lower price point than its Opus models. The model becomes the default for Free and Pro plans, with introductory pricing through August 31.

What happened

Claude Sonnet 5 is the most agentic Sonnet model yet. It can make plans, use tools like browsers and terminals, and run autonomously at a level that previously required larger, more expensive models. Anthropic positions it as a cost-effective alternative to Opus 4.8, OpenAI's GPT-5.6 Sol, and Google's Gemini 3.1 Pro.

Pricing is introductory through August 31: $2 per million input tokens and $10 per million output tokens. After that, reports differ. TechCrunch states $3 per million input and $10 per million output, while Engadget and MacRumors cite $3 and $15 respectively. Opus 4.8 costs $4 per million input and $25 per million output.

Benchmarks cited by Anthropic show Sonnet 5 scoring 63.2% on agentic coding, compared to Opus 4.8's 69.2% and Sonnet 4.6's 58.1%. On knowledge-work tasks, Sonnet 5 slightly outperforms Opus 4.8. Safety improvements include lower rates of undesirable behaviors, better refusal of malicious requests, and reduced hallucination and sycophancy compared to Sonnet 4.6. However, it does not match Opus 4.8 or Mythos on misalignment.

Anthropic also increased rate limits across Chat, Cowork, Claude Code, and the Claude Platform to accommodate higher token usage from agentic tasks. A new tokenizer improves efficiency for heavy agentic users.

Why AI builders should care

Agentic capabilities at a lower price point lower the bar for automating workflows. Builders can now deploy autonomous agents for end-to-end tasks without the cost of Opus-class models. The model's default status on Free and Pro plans means many developers will get it automatically, reducing friction for experimentation.

Safety improvements matter for production deployments. Zapier senior engineer Daniel Shepard noted that Sonnet 5 finished a two-part job end to end that previously stalled halfway. Lovable co-founder Fabian Hedin emphasized that the model refuses unsafe requests cleanly and consistently.

Practical implications

Developers can start experimenting with Sonnet 5 on a budget during the introductory period. The pricing uncertainty after August 31 requires planning for cost management. A comparison of current and projected costs:

Rate limit increases across Claude products support higher usage from agentic workflows. Teams should test Sonnet 5 on tasks like browser automation, code generation, and multi-step planning to gauge cost-performance tradeoffs against Opus 4.8.

Caveats

Post-promo pricing is not uniform across sources. TechCrunch reports $3/$10, while Engadget and MacRumors report $3/$15. Anthropic's official blog post may clarify, but the discrepancy introduces uncertainty for budget planning.

Benchmarks are from Anthropic and tester quotes, not independent verification. Opus 4.8 remains superior on some tasks, and safety gaps persist relative to top-tier Opus and Mythos models. Real-world automation performance depends on integration context and guardrails.

FAQs

What is Claude Sonnet 5 and what can it do?

Claude Sonnet 5 is Anthropic's most agentic mid-size model. It can make plans, use tools like browsers and terminals, and run autonomously at a level that previously required larger, more expensive models. It is the default model for Claude Free and Pro plans at launch.

How is Claude Sonnet 5 priced and what plans include it by default?

Introductory pricing through August 31 is $2 per million input tokens and $10 per million output tokens. After that, reports vary: TechCrunch states $3/$10, while Engadget and MacRumors cite $3/$15. Sonnet 5 is the default model for Free and Pro plans.

How does Claude Sonnet 5 performance compare to Opus 4.8?

On agentic coding, Sonnet 5 scores 63.2% vs Opus 4.8's 69.2%. On knowledge-work tasks, Sonnet 5 slightly outperforms Opus 4.8. Anthropic positions Sonnet 5 as near-Opus performance at lower cost, but Opus 4.8 remains better for high-accuracy tasks and has stronger safety alignment.

What kinds of tasks can Claude Sonnet 5 automate or perform?

Sonnet 5 can plan, use browsers and terminals, and run autonomously. Examples include end-to-end automation like updating Salesforce account tiers and sending enterprise announcements, tasks that previously stalled on earlier models.