Meta's Muse Spark Open reasoning model with 1M-token context arrives on Vercel AI Gateway

What happened

Vercel has added Meta's Muse Spark Open to its AI Gateway, according to a changelog post on the company's blog. The model is an open-weight reasoning model aimed at coding and agentic work, and it can now be reached through the same unified endpoint Vercel customers already use to call other language models.

Vercel says Muse Spark Open handles tasks such as multi-file refactoring, debugging and making sense of large codebases — the kind of work where an agent has to hold a lot of context at once.

A 1M-token window with text and images

The headline specification is a 1 million-token context window. The model accepts both text and image input, which Vercel notes lets an agent work with source code and screenshots inside a single request — useful for pipelines that pair a UI bug screenshot with the relevant files.

The deployment posture is also part of the pitch. According to Vercel, the model is offered with Zero Data Retention (ZDR) and no training on customer data, and developers do not need to create separate provider accounts to use it.

One model name across every interface

The model is addressed as meta/muse-spark-open in all API formats. Per the Vercel blog, that covers:

• The Vercel AI SDK
• The OpenAI-compatible Chat Completions and Responses APIs
• The Anthropic-compatible Messages API
• Coding agents connected to AI Gateway

For tooling such as Claude Code, Codex and Cursor, Vercel describes a short setup path: upgrade the Vercel CLI to the latest version, run vercel ai-gateway coding-agents setup, then select meta/muse-spark-open inside the agent. The model can also be tried out directly in Vercel's model playground.

What the gateway adds

AI Gateway sits as a single API in front of many model providers. Beyond routing requests, Vercel says it handles usage and cost tracking, retries, failover and performance optimizations that the company claims deliver higher uptime than going to the underlying providers directly. Other features include built-in custom reporting, ZDR support, budgets for individual API keys, and routing rules.

Pricing is passed through at provider rates, with no markup and no platform fee on inference — including on Bring Your Own Key (BYOK) requests.

Why it matters

Coding agents are among the most context-hungry applications in AI today, and a 1M-token window that combines code with screenshots maps closely onto how developers actually debug in practice. The open-weight nature of the model matters too: teams that want to avoid vendor lock-in, or that need firm data-retention guarantees, get an open option without having to stand up their own inference infrastructure.

For Vercel, adding Muse Spark Open follows a familiar gateway playbook — every notable model that lands on the unified API makes the product stickier, and pass-through pricing removes one reason to go direct to a provider. For developers already building on Vercel's stack, the practical effect is that a new reasoning model becomes a configuration change rather than an integration project: one model string, compatibility with several major API formats, and immediate availability inside mainstream coding agents.