Muse Wonder on QuickSilver Pro
Muse Wonder is Meta's frontier reasoning model for agentic personal and professional work, with text and image input over a 1M-token context. QuickSilver Pro charges $4.00 input / $20.00 output per million tokens, at OpenRouter list parity; cached input is $0.20 per million tokens.
At a glance
Agentic personal and professional work, coding and long-document reasoning over a 1M-token context.
Pricing comparison ($/1M tokens)
| Provider | Input | Output | vs QSP |
|---|---|---|---|
| QuickSilver Pro | $4.00 | $20.00 | — |
| OpenRouter list price (meta/muse-wonder) | $4.00 | $20.00 | same |
When to use
Choose Muse Wonder for agent workflows, coding and reasoning across large documents or repositories. Its 1M-token context accepts text and images, including screenshots and diagrams. Meta reports, at max effort, SWE-Marathon 53.1%, DeepSWE 78.8% and AutomationBench-AA 39.1%. In Meta's chart, Muse Wonder is ahead of Claude Opus 5.5 and GPT-6 Astra on the first two and behind both on AutomationBench-AA; these are vendor-reported results, not QuickSilver Pro measurements.
When to use something else
For routine chat or tasks with a tight output budget, consider a cheaper model. Reasoning is mandatory and billed as output even though the reasoning text is not returned. max_tokens bounds total output, including reasoning: a verified max_tokens=16 call returned empty content with finish_reason length. Allow enough tokens for both reasoning and the visible answer. Workflows that require forced tool calls or tool_choice none need another model.
Quickstart (curl)
curl https://api.quicksilverpro.io/v1/chat/completions \
-H "Authorization: Bearer $QSP_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "muse-wonder",
"messages": [{"role": "user", "content": "Hello!"}]
}'OpenAI-compatible. One-line migration via base_url.
FAQ
Meta reports SWE-Marathon 53.1%, DeepSWE 78.8% and AutomationBench-AA 39.1% at max effort. These vendor-reported results apply to Muse Wonder and are not independent QuickSilver Pro benchmarks.
No. reasoning: {enabled: false} and reasoning effort none return 400 because reasoning is mandatory. reasoning.effort high is accepted. Reasoning tokens are billed at the output rate and consume max_tokens before the visible answer, but the reasoning text is not returned. The maximum output is 943,718 tokens; a 16-token cap can leave the answer empty.
Use tool_choice auto. required, none and a named function return 400: only auto is supported for tool_choice. Strict response_format json_schema is accepted and returns valid JSON. temperature, top_p and assistant prefill are accepted. Streaming is expected to work as on Muse Spark 1.3.
QuickSilver Pro charges $4.00 input / $20.00 output per million tokens, plus $0.20 per million cached input tokens, matching OpenRouter list. A verified 20-token prompt with 74 completion tokens, including 63 reasoning tokens, cost $0.00156.
Yes. Set base_url=https://api.quicksilverpro.io/v1 and model="muse-wonder" for Chat Completions or Responses. Send text and images and use automatic tool selection; leave mandatory reasoning enabled.