Overview
Claude Sonnet 5, released by Anthropic on June 30, 2026, is the next generation of the Sonnet family and a drop-in upgrade for Claude Sonnet 4.6. It is designed for stronger agentic behavior, coding, tool use, computer-use workflows, and knowledge work at Sonnet-tier pricing.
#Key Features
- Agentic Task Execution: Designed to plan, use tools such as browsers and terminals, and complete multi-step work more reliably than Claude Sonnet 4.6.
- Adaptive Thinking by Default: Requests run with adaptive thinking unless
thinking: {type: "disabled"}is passed; theeffortparameter controls the capability, latency, and token-spend tradeoff. - 1M Context Window: Supports a 1M-token context window by default, with no smaller context variant and no long-context surcharge.
- Broad Platform Availability: Available through the Claude API, Claude Code, Claude Platform on AWS, Amazon Bedrock, Google Cloud, Microsoft Foundry preview, and Claude consumer plans.
#Best Use Cases
- Production Agent Workflows: Multi-step automation, agentic search, tool-heavy reasoning, browser/terminal workflows, and long-running delegated tasks.
- Software Engineering: Coding, debugging, refactoring, code review, test-fix loops, and brownfield repository work where follow-through matters.
- High-Volume Knowledge Work: Research, analysis, structured extraction, business operations, legal research, customer workflows, and internal productivity tools that need a balance of capability and cost.
#Capabilities and Limitations
| Capability | Description |
|---|---|
| Reasoning | Adaptive thinking is on by default; effort supports low, medium, high, xhigh, and max, with high as the default. Anthropic reports stronger agentic, coding, and knowledge-work performance than Sonnet 4.6. |
| Creative Writing | Supports general text generation and long-form writing; Anthropic recommends re-evaluating style prompts because prose style may shift versus earlier Sonnet models. |
| Multimodal | Text and image input, text output, multilingual capabilities, and vision. Native audio/video input or generation is not listed. |
| Response Speed | Listed with "Fast" comparative latency in Anthropic's model table. Lower effort can reduce latency and token usage; higher effort increases thinking/tool-use depth. |
| Context Window | 1M tokens. |
| Max Output | 128K tokens. |
| Tool Use | Supports the same tool and platform feature set as Claude Sonnet 4.6 except Priority Tier is not available; tool use is more readily triggered at higher effort levels. |
| Multilingual | Official docs state multilingual support across current Claude models, but do not publish a separate Sonnet 5 language benchmark. |
#Known Limitations
- Manual extended thinking (
thinking: {type: "enabled", budget_tokens: N}) is removed and returns a 400 error; use adaptive thinking witheffortinstead. - Non-default sampling parameters (
temperature,top_p,top_k) return a 400 error; use system instructions for tone and style control. - Assistant message prefilling remains unsupported and returns a 400 error.
- The new tokenizer produces approximately 30% more tokens for the same text than Claude Sonnet 4.6, so token budgets, context usage, and equivalent-request costs should be remeasured.
- Cybersecurity safeguards are enabled by default; high-risk or prohibited cybersecurity requests may be refused with
stop_reason: "refusal".
Usage example
cURL
curl https://www.namoclaw.ai/api/v1/chat/completions \
-H "Authorization: Bearer $API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "anthropic/claude-sonnet-5-20260630",
"messages": [{ "role": "user", "content": "Hello!" }]
}'Default parameters
| Name | Type | Default | Description |
|---|---|---|---|
| temperature | number | — | — |
| top_p | number | — | — |
| top_k | number | — | — |
| frequency_penalty | number | — | — |
| presence_penalty | number | — | — |
| repetition_penalty | number | — | — |
No reviews yet — be the first to share your experience.