> For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs.empiriolabs.ai/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.empiriolabs.ai/_mcp/server. # Muse Spark 1.3 > Meta's long-horizon agentic reasoning model with a 1,048,576-token context, image, video, and PDF understanding, web search, and tool calling. ![Muse Spark 1.3](https://media.empiriolabs.ai/model-logos/muse-spark-1-1.png) [Meta AI](/providers/meta) · Text Generation `POST /v1/chat/completions` Meta's long-horizon agentic reasoning model with a 1,048,576-token context, image, video, and PDF understanding, web search, and tool calling. ## At a glance | Field | Value | | ------------------- | ---------------------------------------------------------------------------------------------------------------------------- | | Model id | `muse-spark-1-3` | | Model release date | 2026-09-02 | | Input modalities | Text, Image, Video, Document | | Output modalities | Text | | Context window | 1M | | Weight precision | - | | Max output tokens | 131,072 | | Features | reasoning, multimodal, video_understanding, function_calling, structured_output, web_search, cache, document_understanding | | Native inference | No | | New | Yes | | Structured output | JSON Schema | | Supported endpoints | `POST /v1/chat/completions`, `POST /v1/responses`, `POST /v1/messages`, `POST /v1beta/models/muse-spark-1-3:generateContent` | | Alternate model ids | `muse-spark-1.3`, `meta/muse-spark-1-3`, `meta/muse-spark-1.3` | ## Pricing | Charge | Spec | Rate | | ------------------- | -------------------------- | --------- | | Input | per 1M prompt tokens | \$1.25 | | Output | per 1M generated tokens | \$4.25 | | Implicit cache read | per 1M cached input tokens | \$1.00 | | Web search | per search query | \$0.00825 | ## Example request ```bash curl https://api.empiriolabs.ai/v1/chat/completions \ -H 'Authorization: Bearer $EMPIRIOLABS_API_KEY' \ -H 'Content-Type: application/json' \ -d '{"model": "muse-spark-1-3", "messages": [{"role":"user","content":"Hello"}]}' ``` ## Parameters | Parameter | Type | Required | Default | Description | | --------------------- | ------- | -------- | ---------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `max_tokens` | integer | no | `16384` | Maximum number of output tokens to generate. Reasoning tokens count against this budget. · Range: 1 – 131072 | | `temperature` | number | no | `1` | Controls randomness. Lower values make responses more deterministic. · Range: 0 – 2 | | `top_p` | number | no | `1` | Nucleus sampling cutoff. · Range: 0.01 – 1 | | `presence_penalty` | number | no | `0` | Penalizes tokens that already appeared, encouraging new topics. · Range: -2 – 2 | | `frequency_penalty` | number | no | `0` | Penalizes frequent tokens, reducing repetition. · Range: -2 – 2 | | `seed` | integer | no | - | Random seed for more reproducible sampling. | | `reasoning_effort` | enum | no | `"medium"` | Reasoning is always on; this sets how much effort the model spends before answering. Higher effort uses more reasoning tokens. Reasoning text is not returned on chat completions; request reasoning summaries on /v1/responses. · Allowed: `minimal`, `low`, `medium`, `high`, `xhigh` | | `tool_web_search` | boolean | no | false | Enable built-in web search with cited sources. Adds \$0.00825 per executed search query; a single request can run more than one search. | | `tools` | array | no | `[]` | OpenAI-compatible function and custom tool definitions. On /v1/responses, tool_search and defer_loading can discover deferred tools. | | `tool_choice` | object | no | - | OpenAI-compatible tool choice control. This model supports auto and none. | | `parallel_tool_calls` | boolean | no | true | Allow the model to request multiple function tools in one response. | | `response_format` | enum | no | - | Return structured JSON output. JSON mode returns any valid JSON object; JSON Schema mode enforces an exact schema. | ## Notes Reasoning is always on and cannot be disabled. reasoning_effort accepts minimal through xhigh. On /v1/chat/completions the reasoning text is not returned; on /v1/responses you can request reasoning summaries with reasoning.summary, and reasoning tokens are billed as output tokens and count against the output budget either way. Text, image, MP4 video, and PDF inputs are supported; up to 50 images can be attached. Audio input is not supported on this checkpoint; the Muse Spark 1.1 and 1.2 checkpoints accept audio. PDFs use text from the first 100 pages and page images from the first 50 pages. Built-in web search adds \$0.00825 per executed search query and reports the count in usage.tool_usage. OpenAI Responses requests also support custom tools and deferred tool search. tool_choice supports auto and none. Prompts and completions sent to this Standard checkpoint are not used by Meta to train its models. **Per-tool billing (`usage.tool_usage`)** When this model invokes tools inside a single request, the response carries a normalized `usage.tool_usage` map alongside the token counts. The example below shows the shape. Exact field names, units, and which tools appear can vary by tool: ```json "usage": { "prompt_tokens": 123, "completion_tokens": 456, "cost_usd": 0.0042, "tool_usage": {"web_search": 3} } ``` The tool counts are already factored into `cost_usd` and are surfaced so you can audit per-tool billing. The field is omitted when no billed tool was invoked. --- *Machine-readable schema:* `GET https://api.empiriolabs.ai/v1/models/muse-spark-1-3`.