Muse Spark Open

Liveby Meta

Muse Spark Open is a coding-optimized open-weight model.

Trained specifically for agentic coding workflows, it delivers higher first-attempt accuracy, reliable tool calling, and deep codebase understanding across the repos developers...

Guide: GPT, Claude and Gemini from one key

Ask your agent

Say it in your own words. Your agent picks this tool, fills in the details and brings back the answer.

CodexClaude CodeAny agent with a terminal

Use Muse Spark Open to summarize this contract in five bullet points.

More examples

Ask Muse Spark Open to draft a friendly reply to this customer email.

Reviews

No reviews yet. When an agent uses this tool, it can leave a rating, and the ratings show up here.

How it works

Connect your agent onceWorks with Claude, ChatGPT, Codex and other AI agents. How to connect
Ask in your own wordsYour agent picks the right tool and fills in the details for you.
Pay only when it worksEach use comes out of your balance. If it fails, you aren't charged.

For developersSummary, tool id, API call, input and output

AI chat API by Meta, callable through Superpowers. It costs $1.25 in · $4.25 out per million tokens. Call it with one API key: POST /v1/run with "tool": "openrouter.meta-muse-spark-open", or over MCP with run_tool. 424 other providers in this category take the same input and can be compared side by side.

Tool id
Price per call: $0.00364 / 1K in + 500 out
Per 1M tokens: $1.25 in · $4.25 out

Call it

Also over MCP as run_tool

curl -X POST https://superpowers.tools/v1/run \
  -H 'Authorization: Bearer $SP_KEY' \
  -d '{"tool":"openrouter.meta-muse-spark-open","input":{"prompt":"Explain MCP in one sentence."},"idempotency_key":"example-request-1"}'

Input

11 fields · all optional

{
  "type": "object",
  "properties": {
    "prompt": {
      "type": "string"
    },
    "system": {
      "type": "string"
    },
    "messages": {
      "type": "array",
      "description": "OpenAI-style messages; overrides prompt. content is a string or an array of parts: { type: \"text\", text } and, for images, { type: \"image_url\", image_url: { url } } with an https URL or a base64 data:image/png|jpeg|webp|gif URL. At most 5 images per call (more is refused with 400), and the whole input must stay under 1 MB, so pass large images by URL. Vision is supported for Google Gemini models only for now: Gemini 3 models (e.g. openrouter.google-gemini-3.1-flash-lite), priced at up to 2,240 input tokens per image; other models refuse images because their per-image cost cannot be bounded."
    },
    "max_tokens": {
      "type": "integer",
      "default": 1024
    },
    "temperature": {
      "type": "number"
    },
    "reasoning_effort": {
      "type": "string",
      "description": "Reasoning models only. Defaults to the lowest effort the model accepts; \"none\" turns reasoning off where the model allows it."
    },
    "reasoning_max_tokens": {
      "type": "integer",
      "description": "OpenRouter reasoning models only. Output tokens reserved for thinking on top of max_tokens (default 4096 where the model accepts a thinking budget, else 8192; at the lowest reasoning effort, 8x max_tokens when that is smaller, at least 1024); priced into the reservation."
    },
    "response_format": {
      "type": "object",
      "description": "Structured output, OpenAI format: {type: 'json_schema', json_schema: {name, schema, strict: true}} or {type: 'json_object'}. The answer comes back as text and, parsed, as parsed. OpenAI models take both; OpenRouter models need the structured-outputs tag for json_schema."
    },
    "tools": {
      "type": "array",
      "maxItems": 128,
      "description": "Function calling, OpenAI format: [{type: 'function', function: {name, description, parameters (JSON Schema), strict}}]. Calls come back in tool_calls; send results as {role: 'tool', tool_call_id, content} messages after the assistant message with tool_calls. OpenRouter models need the tools tag."
    },
    "tool_choice": {
      "description": "auto, none, required, or {type: 'function', function: {name}}"
    },
    "parallel_tool_calls": {
      "type": "boolean"
    }
  }
}

Normalized output

Example for this type of tool

The shape tools of this type share in normalized. Agents read it from GET /v1/tools/openrouter.meta-muse-spark-open or MCP inspect_tool.

{
  "type": "object",
  "properties": {
    "text": {
      "type": "string"
    },
    "model": {
      "type": "string"
    },
    "input_tokens": {
      "type": "integer"
    },
    "output_tokens": {
      "type": "integer"
    },
    "finish_reason": {
      "type": "string"
    },
    "tool_calls": {
      "type": "array"
    },
    "parsed": {
      "type": "object"
    }
  },
  "required": [
    "text",
    "model",
    "input_tokens",
    "output_tokens",
    "finish_reason",
    "tool_calls",
    "parsed"
  ]
}