Choose a model

Muse Wonder, Muse Spark, and Muse Ember are reasoning models that take the same requests on the same endpoints, so moving a workload between them is a model ID change. Pick one by what each part of your app needs — or run them together in the same agent.

Compare the models

Muse WonderMuse SparkMuse Ember
Built forComplex software engineering and long-horizon agentic workAgentic and coding work at scaleEfficient execution of high-volume, repeated work
Use it whenCapability matters more than cost per tokenThe main loop or a harder task needs strong reasoning at a lower cost than Muse WonderA larger model plans, and sub-agents run many bounded tasks
Model IDsmuse-wonder-1.0muse-spark-1.3 (recommended), muse-spark-1.2, muse-spark-1.1muse-ember-1.0
InputText, image, video, audio, and PDFText, image, video, audio, and PDFText, image, video, audio, and PDF
Context window1,048,576 tokens1,048,576 tokens1,048,576 tokens
Reasoning effort"minimal" to "max"; default "max""minimal" to "max" on muse-spark-1.3; default "high""minimal" to "max"; default "max"
Contributor mode on Model APINot availablemuse-spark-1.3-contributor, muse-spark-1.2-contributorNot available
PricingMuse Wonder pricingMuse Spark pricingMuse Ember pricing, with long-context rates over 200K input tokens
AvailabilityMuse Wonder availabilityMuse Spark availabilityMuse Ember availability

Choose by workload

WorkloadModel
Multi-file coding changes in a real repositoryMuse Wonder
Multi-hour agent runs and sub-agent orchestrationMuse Wonder
The main loop of a coding agent or agent frameworkMuse Wonder
Computer useMuse Wonder, or Muse Spark or Muse Ember at lower cost
A lower-cost main loop for agents and assistantsMuse Spark
Knowledge-heavy questions answered without retrievalMuse Spark or Muse Wonder
Sub-agent tasks handed out by a main agent, including coding subtasksMuse Ember
Tool calls, tight tool loops, and latency-sensitive turnsMuse Ember
Classification, routing, and other high-volume requestsMuse Ember
Running the same task over thousands of images, files, or recordsMuse Ember
Discounted prices on Model API in Contributor modeMuse Spark, through a Contributor-mode variant
Audio understandingMuse Spark 1.2
Speech to textMuse Voice Transcribe
Image generation and editingMuse Image
Image and video segmentationSegment Anything Model

Combine models in one agent

Most agents don't need one model for everything. Put Muse Wonder on the main loop, where planning and multi-step reasoning decide the outcome, or Muse Spark when the main loop needs to cost less. Put Muse Ember on sub-agents and background tasks, which run many short, focused jobs where cost adds up: the larger model plans, and Muse Ember handles the volume.

The harnesses you already use support this split:

  • Claude Code and the Claude Agent SDK: map the opus alias to muse-wonder-1.0, the sonnet alias to muse-spark-1.3, and the haiku alias and sub-agent model to muse-ember-1.0. See Set up Claude Code and Build with the Claude Agent SDK.
  • Codex: set muse-wonder-1.0 as the default model, and switch to muse-spark-1.3 for faster, lower-cost sessions. See Set up Codex.
  • Your own agent loop: pass a different model on each request. The reasoning models accept the same request body; Switch between models lists the few settings that differ.

Switch between models

Moving a workload from one model to another is a model ID change. Keep these differences in mind:

  • Reasoning effort: muse-wonder-1.0, muse-spark-1.3, and muse-ember-1.0 accept "minimal" to "max". "max" isn't available on muse-spark-1.2, muse-spark-1.1, or muse-spark-1.3-contributor. When a request sets no effort, Muse Wonder and Muse Ember use "max" and Muse Spark uses "high"; see Reasoning.
  • Contributor mode: on Model API, only Muse Spark has Contributor-mode variants. Muse Wonder and Muse Ember are Standard mode only; see Muse Wonder availability and Muse Ember availability.
  • Muse Wonder Full: the higher-capability variant of Muse Wonder is available only in Muse Code, so no model ID on Model API selects it; see Muse Wonder availability.
  • Audio input: response quality for audio on Muse Spark 1.3 may be degraded. Use Muse Spark 1.2 or Muse Voice Transcribe instead.
  • Long-context pricing: Muse Wonder and Muse Spark charge the same rate at any context length. Muse Ember bills a request with more than 200K input tokens at long-context rates, so compare costs at your real input sizes; see Muse Ember pricing.
  • Harder work: for hard single-shot coding problems, knowledge-heavy questions, or visual coding, use Muse Spark or Muse Wonder rather than Muse Ember.

Next steps