Muse Spark 1.3 API & Pricing: Model IDs, Costs and Setup

01

Muse Spark 1.3 model IDs on Vercel

Vercel AI Gateway currently exposes the standard route as meta/muse-spark-1.3 and the lower-cost contributor route as meta/muse-spark-1.3-contributor.

Those are Vercel gateway identifiers. Meta’s direct Model API can use provider-specific endpoint conventions, so do not assume a gateway model string is universal across providers.

02

Standard Vercel AI Gateway pricing

Vercel currently lists meta/muse-spark-1.3 at $1.25 per million input tokens, $4.25 per million output tokens and $0.15 per million cached-input reads.

The Vercel provider listing shows a 1M shared context window and a 1M maximum output limit for the route at publication time. Treat both limits and prices as provider-specific operational data that can change.

03

Contributor pricing and the data-use tradeoff

Vercel lists the contributor route at $0.10/M input, $0.20/M output and $0.002/M cached-input reads. It is dramatically cheaper than the standard route for the same broad model family.

Vercel describes contributor usage as permitted to improve Meta’s products. That makes the route a policy decision as well as a pricing decision; sensitive source code, private client data, credentials and regulated information should not be sent without explicit review of the current terms.

04

Context window and supported inputs

Vercel documents a 1 million-token context window with text, image and PDF inputs. That is useful for large repositories and document-heavy agents, but prompt and response share the available context budget.

Before relying on very large prompts, benchmark retrieval quality, latency and total spend against a smaller-context strategy using summaries, retrieval or cached repeated material.

05

How to plan cost for agent workflows

Agent cost depends on more than base token rates. Measure repeated context, cache hit rate, tool-call loops, retries, output length and successful-task rate. A lower token price can still be expensive if the workflow repeatedly fails or re-reads the same context uncached.

Meta says 1.3 reduces unnecessary turns and tokens versus 1.2, but that remains a vendor claim. Run a fixed evaluation set on your own coding or business tasks before migrating production traffic.

06

Reasoning status and deployment checks

Meta says existing reasoning modes are available now, while max reasoning is coming after additional safety testing. Do not build a production dependency on max reasoning until it is actually enabled on the surface you use.

• Recheck current provider prices and cache rates
• Confirm context and output limits
• Review contributor data-use terms
• Check regional access and rate limits
• Set retry and budget ceilings
• Keep secrets out of prompts