Vercel Adds Muse Spark 1.3 to AI Gateway Model Lineup.

Vercel has added Muse Spark 1.3, Meta’s latest agentic and coding model, to AI Gateway. The release gives developers a new option for long-context workflows, with support for text, images, and PDFs through a single model endpoint.

According to Vercel’s changelog announcement, Muse Spark 1.3 is available in both the standard and contributor pricing tiers. Developers can call it using the model identifier meta/muse-spark-1.3.

The addition is notable for its 1 million-token context window, which allows an application to send substantially larger collections of source code, documentation, transcripts, or other material in a single request. Vercel also says the model improves on previous Muse Spark releases when completing agent tasks and writing code.

Muse Spark 1.3 is designed to accept three input formats:

• Text, including prompts, documents, and source code
• Images, for workflows that need visual context
• PDFs, such as manuals, reports, and specifications

That combination could make the model useful for applications that need to reason across mixed project materials. A coding agent, for example, could review a large repository alongside API documentation and a PDF migration guide. A support or operations tool could analyze written instructions together with screenshots supplied by a user.

Vercel says Muse Spark 1.3 requires fewer turns for coding tasks and produces less filler than the previous release. Fewer turns matter in agentic workflows because each additional model exchange can increase latency and consumption. Less filler may also make generated patches, explanations, and tool calls easier for developers to inspect.

The company has not published comparative benchmark results in the announcement, so builders will need to evaluate those improvements against their own repositories and prompts. Tests should focus on whether the model completes multi-step tasks reliably, uses available context effectively, and produces changes that pass existing checks.

Developers using Vercel’s AI SDK can select the model by passing its identifier to streamText:

import { streamText } from 'ai';

const result = streamText({
  model: 'meta/muse-spark-1.3',
  prompt: 'Migrate this service off the deprecated API.',
});

The model is also presented through AI Gateway options for API access, Claude Code, Codex, Hermes, and Chat Completions. That gives teams several ways to introduce it into existing development and agent workflows without treating it as a separate infrastructure deployment.

For small businesses and independent builders, the practical advantage is optionality. A team can test Muse Spark 1.3 for a document-heavy assistant or coding workflow while retaining the gateway as its integration layer. It can then compare the model’s responses, latency, and usage with other available models before making it a default.

This release also continues Vercel’s broader expansion of AI Gateway. It follows another recent model addition and the introduction of per-user AI Gateway budgets, which give teams more control over individual usage. Together, those changes position the gateway as both a model access layer and a place to manage how AI capacity is consumed across an application.

Vercel has not provided token rates in the announcement. Prospective users should consult the model details and gateway pricing pages before moving a high-volume workload into production.

What is Vercel Muse Spark 1.3?

Muse Spark 1.3 is a Meta AI model available through Vercel AI Gateway. It is intended for agentic and coding tasks, provides a 1 million-token context window, and accepts text, image, and PDF inputs.

How can developers access Muse Spark 1.3?

Developers can set the model identifier to meta/muse-spark-1.3 in an AI Gateway request. Vercel’s example uses the AI SDK’s streamText function, while the model page also lists API, Claude Code, Codex, Hermes, and Chat Completions options.

Which Vercel AI Gateway tiers include Muse Spark 1.3?

Vercel says Muse Spark 1.3 is available in both its standard and contributor pricing tiers. The announcement does not specify token rates, so teams should check Vercel’s current model details before estimating production costs.