LLM & ChatActive
claude / opus-5
opus-5
Claude Opus 5 is a premium text model with vision for agentic coding and document work. Upload images or PDFs and get long, detailed answers.
Partner LLM
Model ID
opus-5
Provider
claude
Updated
1787229096
wiro playground—claude/opus-5
Updated 1787229096
Overview
Opus 5 is Claude’s Opus-tier model developed by Anthropic. It takes text plus optional images or PDFs, then writes a single, high-detail text answer. It uses adaptive thinking and an effort control so you can trade speed and token spend for deeper reasoning. It shines when you need one model to read long material, keep context, and produce work you can act on.
What you can build
- Repository-level coding help like design plans, refactors, and multi-file change guidance
- Debugging and root-cause analysis from logs, error traces, and screenshots
- Document Q&A for PDFs like contracts, financial reports, and policy handbooks
- Extraction of tables and key figures from PDF pages into clean summaries
- Due diligence checklists and risk memos from mixed sources and internal docs
- Long-form drafting like technical specs, SOPs, and decision records
Inputs
- A required instruction describing what you want, written as plain text.
- Optional files you want the model to read: - Images: JPEG, PNG, GIF, or WebP.
- Documents: PDF only.
- For PDFs, use standard, unencrypted PDFs. Keep documents under 100 pages per request. Keep the full request payload under 32 MB.
- Optional system-level guidance that sets rules for tone, format, and boundaries.
- Optional conversation identifiers that let the app keep chat history across turns.
- An optional effort level choice (low, medium, high, extra-high, or max). Higher effort spends more tokens on reasoning.
- An optional web search toggle. When enabled, the model can pull current web sources and cite them.
- An optional randomness control from 0.0 to 1.0. Higher values vary wording more.
- An optional cap on response length, up to 128,000 tokens.
Outputs
The model returns a text response.
- For coding tasks, the response often includes code blocks, diffs, and step-by-step plans.
- For documents, the response summarizes key points and can answer targeted questions from the PDF.
- When web search is enabled, the response can include source citations tied to the retrieved pages.
Recommended settings
- Keep effort at High for most serious work. Move to Extra-high for demanding coding and agent-style tasks.
- Use Max effort only when you want unconstrained depth and can afford higher token use.
- When you run Extra-high or Max effort, allow a large output budget. Start at about 64,000 tokens, then tune.
Limitations
- Opus 5 has a reliable knowledge cutoff (May 2026). Use web search for newer facts.
- At Extra-high or Max effort, you can’t disable the model’s thinking mode.
- Some higher-risk offensive cybersecurity requests may trigger a safety fallback to a different model.
- PDF understanding depends on input quality. Scanned pages, low-contrast text, rotated pages, and messy tables can reduce accuracy.
- Images with small text, blur, or heavy compression can lead to missed details.
Safety & compliance
- The model applies safety checks and may refuse disallowed requests.
- Treat outputs as assistant-written drafts. Review before you ship code, legal text, or security guidance.
- If you enable web search, keep citations with the output when you show results to end users.
- Don’t upload secrets or personal data you don’t have rights to share.
API quick start
Run opus-5 with a single API call.
POST https://api.wiro.ai/v1/Run/claude/opus-5
{
"prompt": "Tell me about Türkiye",
"inputAll": "https://your-cdn.com/input.png",
"user_id": "...",
"session_id": "..."
}