Back to models
Chat Models

Meta

Muse Spark 1.2

Meta’s previous Muse Spark version supports text, image, video, audio, and PDF inputs for long-context reasoning and tool-assisted work.

MultimodalReasoning1M context
Open AppBrowse All Models

At a glance

Know the model before you prompt.

Meta API specifications
Context window
1,048,576 tokens
Maximum output
Not verified
Inputs → output
Text + Images + Video + Audio + PDF → Text
Knowledge cutoff
Not verified

API model ID: muse-spark-1.2

Capabilities & boundaries

What it supports. Where the limits are.

Tool support requires the appropriate API integration; a supported tool is not automatically active in every chat.

Supported API features and tools

  • Long-context multimodal understanding
  • Function/tool calling and structured outputs
  • Web-search grounding through an enabled tool
  • Responses, Chat Completions, and Messages API compatibility
  • Prompt caching
  • Responses API continuation with previous_response_id

Before you choose

  • The max reasoning setting is not supported by this version; it is specific to Standard-tier 1.3.
  • Muse Spark outputs text, not generated images or speech; Meta has separate model families for those outputs.
  • Reasoning cannot be disabled with none, and log probabilities are not supported.
  • No maximum output-token limit or training cutoff is established by the reviewed model documentation.

Choose the reasoning effort

minimallowmediumhighxhigh

Choose minimal, low, medium, high, or xhigh. Omitting the effort parameter uses a model-determined level. Do not send none or assume 1.3’s max setting works on this version. Evaluate the lowest effort that meets your quality checks.

Unsupported settings: none, max.

  • Reasoning tokens and visible answer tokens share the output budget and are both billed as output.
  • Use reasoning.effort with Responses or reasoning_effort with Chat Completions. API compatibility does not guarantee every client feature is supported.
  • For multi-turn reasoning continuity, use Responses with previous_response_id or the documented encrypted reasoning replay. External Chat Completions callers do not receive replayable private reasoning.
  • A separate Contributor model ID is available, with different prices and permission to train on prompts and completions. Do not assume EZ Ai Assist uses or exposes that tier.

Put it to work

Start with a more useful prompt.

Original examples from EZ Ai Assist. Adapt these to your task and the features available in your workspace.

Workflow 01

Summarize an authorized recording

Use audio only where the integration supports it.

Summarize the recording I provide into topics, decisions, and unresolved questions. Include timestamps for important statements when available. Mark uncertain transcription explicitly and do not infer speaker identities. Extract commitments only when they are clearly stated; leave absent owners or deadlines unassigned.

Workflow 02

Reconcile a PDF with a presentation

Check consistency across supplied multimodal material.

Compare the supplied report PDF with the presentation screenshots. List mismatched numbers, omitted caveats, and claims whose sources are unclear. Cite page or slide numbers for each finding. Do not silently correct either document; return a review checklist for the author.

Workflow 03

Review a small tool transcript

Find evidence-backed failures in an existing run.

Review this tool-use transcript against the requested task. Identify the first unsupported assumption, failed validation, or unnecessary action. Quote the relevant step and suggest the smallest change that would prevent recurrence. Separate tool failures from planning errors and propose one regression test for each confirmed issue.

Developer reference

Meta API pricing

These are Meta Standard API reference prices, not EZ Ai Assist subscription prices.

View EZ Ai Assist plans
Standard processing · USD per 1,000,000 tokens
Token typePrice
Input$1.25
Cached input$0.15
Output (including reasoning)$4.25
Additional API fees · USD
UsageRate and unit
Web-search grounding$2.50per 1,000 search queries
  • Meta lists the same Standard token rates for Muse Spark 1.3, 1.2, and 1.1, with no long-context premium.
  • Standard API prompts and completions are not used to train Meta models according to Meta’s pricing documentation. This statement is about direct Meta API terms, not a claim about EZ Ai Assist’s integration.
  • The separate muse-spark-1.2-contributor ID costs $0.10 input, $0.002 cached input, and $0.20 output per 1M tokens. This discount permits training on your prompts and completions; it is not the Standard tier.
  • Web-search queries add charges beyond token costs; one request can use more than one query. Confirm current rates and inspect response usage before estimating spend.

Common questions

A few things worth knowing.

When should I use Muse Spark 1.2?

Evaluate Muse Spark 1.2 for established multimodal workflows or audio understanding while comparing its results with 1.3 on non-audio tasks. It shares the documented context size and Standard prices, but does not support 1.3’s max effort setting.

What can this model take as input?

Meta lists text, image, video, audio, and PDF inputs, with text output. Your integration determines which inputs can actually be attached. Understanding an image or recording does not imply image or speech generation.

How does reasoning affect price and latency?

Reasoning tokens count toward the output budget and are billed at the output rate. Higher effort can take longer and leave less room for visible text within the same budget. Muse Spark does not support turning reasoning off with none.

Can I use the max effort setting?

No. Meta documents max only for Standard-tier Muse Spark 1.3. This version supports minimal through xhigh; a missing parameter uses model-determined reasoning rather than a named default.

How is Contributor different from Standard?

Contributor is a separate discounted API model ID that permits Meta to train on prompts and completions. Standard has different rates and data-use terms. This guide does not assume your EZ Ai Assist workspace uses either specific tier.

Is Muse Spark an open-weight Llama model?

This guide covers Muse Spark hosted on Meta Model API. Meta’s model overview separately identifies its open-weight offerings; do not assume Muse Spark’s weights, deployment options, or licensing match Llama or another family.

Does this page confirm all these features are available in EZ Ai Assist?

No. It is a provider API reference with original workflow examples. Check the app for model access, input support, tools, reasoning controls, and applicable data terms. EZ Ai Assist plan prices are separate from Meta’s per-token charges.

Check the source

Official documentation

Specifications and API prices checked on . Example prompts and workflow advice are editorial guidance from EZ Ai Assist.

Same provider

More from Meta

Meta

Muse Spark 1.3

Meta’s latest Muse Spark for coding and multi-step tool workflows, with a 1M-token context and max reasoning on the Standard tier.

Meta

Muse Spark 1.1

Meta’s original Muse Spark version supports multimodal understanding and reasoning with a 1M-token context and Standard-tier API pricing.