Day 0 support for Meta Muse Spark 1.3 on LiteLLM!

[misbahsy]

LiteLLM now offers muse-spark-1.3 and muse-spark-1.3-contributor on day 0, through the meta/ route on the Meta Model API. Meta ships 1.3 today in Muse Code and the Model API, tuned for long-horizon agentic and coding work, and reports roughly 20% fewer tool calls and 25% fewer tokens than 1.2 internally.

What's new:

Standard pricing is unchanged from 1.2 and 1.1 - $1.25 / MTok input, $4.25 / MTok output, $0.15 / MTok cached input.

The contributor tier offers a 21x output discount - meta/muse-spark-1.3-contributor runs $0.10 / $0.20 / $0.002, and Meta's terms are that your prompts and completions may be used to train future models, where the standard tier is never trained on. Rate limits drop to 100 RPM from 3,000, enforced per team rather than per key.

1M context, 128K output - a 1,048,576-token context window and a 131,072-token max output, with text, image, video and PDF input, function calling, prompt caching, and web search grounding at $2.50 per 1,000 queries. reasoning_effort accepts minimal through xhigh.

Get started, no redeploy needed:

Muse Spark 1.3 is in the LiteLLM model cost map as of PR #39417. Reload your pricing in the UI under Models + Endpoints -> Price Data -> Reload Price Data (or POST /reload/model_cost_map as an admin) and add the following model entry.

model_list:
  - model_name: muse-spark-1.3
    litellm_params:
      model: meta/muse-spark-1.3
      api_key: os.environ/META_API_KEY

  # Cheaper tier, but Meta may train on your prompts and completions
  - model_name: muse-spark-1.3-contributor
    litellm_params:
      model: meta/muse-spark-1.3-contributor
      api_key: os.environ/META_API_KEY

Read the full guide → Day 0 support: Meta Muse Spark 1.3

Misbah Syed
DevRel @ LiteLLM