Muse Spark Open on QuickSilver Pro
Muse Spark Open is Meta's open-weight model for coding and agent workflows, with a 1M-token context. OpenRouter list parity: $1.25 input / $4.25 output per million tokens; cached input is $0.15.
At a glance
Coding agents, tools, structured output and image understanding.
Pricing comparison ($/1M tokens)
| Provider | Input | Output | vs QSP |
|---|---|---|---|
| QuickSilver Pro | $1.25 | $4.25 | — |
| OpenRouter list price (meta/muse-spark-open) | $1.25 | $4.25 | same |
When to use
Use for coding and agent workflows that need text and image input, automatic tools, forced named functions or strict JSON Schema output. These features and streaming with usage were verified on both pinned providers.
When to use something else
Choose another model when you need reasoning disabled. Mandatory reasoning can consume a small output budget before a visible answer appears. Audio, video and file input have not been validated on this route.
Quickstart (curl)
curl https://api.quicksilverpro.io/v1/chat/completions \
-H "Authorization: Bearer $QSP_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "muse-spark-open",
"messages": [{"role": "user", "content": "Hello!"}]
}'OpenAI-compatible. One-line migration via base_url.
FAQ
No. Reasoning is enabled by default; upstream lists medium effort and rejects reasoning: {enabled: false}. reasoning: {effort: "low"} worked on both providers. Use enough output tokens for reasoning and the final answer; low effort helped a schema response finish within the probe budget.
$1.25 input / $4.25 output per million tokens, plus $0.15 per million cached input tokens. These match OpenRouter list prices at factor 1.0. Observed usage.cost matched the list rates, including Together cache hits; Modal cache pricing is listed upstream but a cache hit was not observed.
The route pins Modal, then Together, with zero data retention and no fallback outside that order. Both passed tools, forced functions, JSON Schema, images and streaming. Common advertised limits are 1M context tokens and 262K output tokens; probes used at most 256. Both providers report unknown quantization. Meta publishes 816B total / about 42B active parameters under the custom Muse Model License.
Use base_url=https://api.quicksilverpro.io/v1 and model="muse-spark-open" with an OpenAI-compatible client. Chat Completions and Responses share the catalog route.