Meta's Muse Spark 1.3 arrives on Vercel AI Gateway with 1M token context

Meta's latest model lands on AI Gateway

Vercel has announced that Muse Spark 1.3, Meta's newest language model, is now available through AI Gateway, the company's routing layer for accessing models from different providers. According to the Vercel blog, the release ships with two pricing options: a standard tier and a cheaper contributor tier.

One million tokens of context, plus text, image and PDF input

Vercel says the model supports a context window of one million tokens and accepts text, images and PDFs as input. The company positions the release as an upgrade over earlier Muse Spark versions in two areas in particular: agent work and coding.

For programming specifically, Vercel reports that Muse Spark 1.3 gets to working solutions in fewer conversational turns and pads its output with less filler than the previous release. Fewer turns matter for agent-driven workflows, where each round trip costs time and tokens, and leaner output is easier to review and maintain.

The contributor tier trades data for lower prices

The more unusual detail in the announcement is how the contributor tier works. It is not a separate model: Vercel states it uses the same weights, the same capabilities and the same context window as the standard tier. The distinction is what happens to your traffic. Inputs and outputs sent to the contributor tier are used by Meta to train and improve its models, and in exchange the pricing is lower.

For teams with strict data policies, the standard tier keeps traffic out of Meta's training pipeline; for everyone else, the contributor option offers a discount in exchange for allowing that usage. Vercel notes that per-million-token rates are unchanged from Muse Spark 1.2 on both tiers, so existing users migrating from the previous version should not see a price difference.

Wiring it into coding agents

To use the model, developers set the model identifier to meta/muse-spark-1.3. Because the release is aimed partly at coding agents, Vercel points to its coding agents guide and provides a setup command, vercel ai-gateway coding-agents setup, which connects the gateway to tools including Claude Code, Codex and Cursor. Once configured, Muse Spark 1.3 can be selected directly inside the agent. The model is also available to try in Vercel's model playground alongside the other language models offered through AI Gateway.

Why it matters

Two things stand out about this release. First, the combination of a one-million-token context window, multimodal input and claimed improvements in agentic coding puts Muse Spark 1.3 in direct competition with the frontier models that developers currently reach for in tools like Cursor and Claude Code. Making it available through AI Gateway means teams already routing requests through Vercel can swap it in with a one-line model change, lowering the barrier to evaluating it against incumbents.

Second, the contributor tier is a data-for-discounts arrangement made explicit at the API level rather than buried in terms of service. That transparency lets teams make a deliberate choice between cheaper inference and keeping their prompts and code out of Meta's training data — a decision that, depending on the sensitivity of the codebase, may be worth more than the discount itself.