LLM & ChatActive
claude / sonnet-5
sonnet-5
Claude Sonnet 5 is an agent-ready text model for coding, reasoning, and long-context work. It can read images and PDFs and returns text answers.
Partner LLM
Model ID
sonnet-5
Provider
claude
Updated
1787229101
wiro playground—claude/sonnet-5
Updated 1787229101
Overview
Claude Sonnet 5 is a multimodal text model made by Anthropic and shipped under Claude. It reads long text plus optional images and PDFs, then writes a text response you can use directly. Adaptive thinking is on by default, so it can plan before answering. The 1M-token context window helps on large repos and long documents.
What you can build
- Code review and refactors across large codebases, with long multi-step plans
- Agent-style workflows that call tools like web search or code execution when available
- PDF Q&A for reports, contracts, manuals, and slide-style documents
- Table and chart interpretation from PDFs, including “explain this figure” requests
- Screenshot and diagram understanding from images, like UI audits and flowcharts
- Structured extraction drafts, like invoice fields, clause lists, and risk summaries
Inputs
- A required task prompt as plain text. Ask for a specific output format when you need it.
- Optional file attachments for context. Supported types include JPEG, PNG, GIF, and WebP images, plus PDF documents.
- Optional system-level instructions that set rules for tone, role, and boundaries.
- An optional thinking effort level, from low to max, to trade off depth vs brevity.
- An optional web search toggle, when you need up-to-date facts outside the model’s cutoff.
- Optional user and session identifiers to keep multi-turn chat history consistent.
- An optional response length cap. Claude Sonnet 5 supports up to 128,000 output tokens.
Outputs
- A text response that typically works well as Markdown.
- For disallowed or high-risk requests, the model may return a refusal instead of an answer.
- When your workspace enables tool use, the model can return tool-call style actions alongside text.
Recommended settings
- Use high thinking effort for most coding and agent tasks.
- Use xhigh or max thinking effort for the hardest multi-step debugging or planning.
- Use low thinking effort for quick summaries and simple rewrites.
- Keep sampling settings at the platform defaults. Claude Sonnet 5 rejects non-default sampling on the native Claude Platform.
- Set a generous output cap for long reports. Tight caps can truncate answers sooner than expected.
Limitations
- Claude Sonnet 5 uses a new tokenizer. The same text can consume about 30% more tokens than Sonnet 4.6.
- The model supports a 1M-token context window, but dense PDFs can still fill it early.
- PDF requests have hard constraints. The full request payload must fit within a 32 MB limit, and PDFs must be standard and not encrypted.
- Very long PDFs may need splitting. The platform limit is up to 600 pages per request.
- Scanned PDFs, low-resolution screenshots, heavy compression, and messy layouts reduce accuracy.
- Some API features from older Sonnet variants are not supported. Assistant-message prefilling can be rejected.
Safety & compliance
- Follow Claude’s content and safety policies for disallowed content.
- Claude Sonnet 5 includes real-time cybersecurity safeguards. It may refuse risky security requests.
- Treat content from web pages and uploaded files as untrusted. Prompt injection is possible.
- Don’t upload sensitive personal data unless you have rights and a clear need.
API quick start
Run sonnet-5 with a single API call.
POST https://api.wiro.ai/v1/Run/claude/sonnet-5
{
"prompt": "Tell me about Türkiye",
"inputAll": "https://your-cdn.com/input.png",
"user_id": "...",
"session_id": "..."
}