Sign InOpen Brain
VercelEngineering PostOfficial Source

Muse Spark 1.3 now available on AI Gateway

Muse Spark 1.3 gives coding agents a 1M-token, multimodal model through Vercel, with a cheaper contributor tier that permits Meta to train on submitted inputs and outputs.

Vercel · Sep 2, 2026
Open Source Open MarkdownOpen JSON
Source Summary

Meta’s **Muse Spark 1.3** is available through Vercel AI Gateway with a **1M-token context window** and text, image, and PDF input. Vercel says it uses fewer coding turns and produces less filler than its predecessor.

Practical Implication

Builders can select meta/muse-spark-1.3 in Claude Code, Codex, Cursor, and other connected agents. The contributor tier uses the same model and drops per-million-token rates to **$0.10 input and $0.20 output**, versus $1.25 and $4.25 on the standard tier.

Agent-Ready Context
Meta’s **Muse Spark 1.3** is available through Vercel AI Gateway with a **1M-token context window** and text, image, and PDF input. Vercel says it uses fewer coding turns and produces less filler than its predecessor.

Builders can select meta/muse-spark-1.3 in Claude Code, Codex, Cursor, and other connected agents. The contributor tier uses the same model and drops per-million-token rates to **$0.10 input and $0.20 output**, versus $1.25 and $4.25 on the standard tier.

The discount carries a data tradeoff: Meta may use contributor-tier inputs and outputs for model training. The material provides no benchmark results for the claimed coding improvements, and both tiers retain Muse Spark 1.2 pricing.
Connected Context · Feed7 Judgment

Muse Spark 1.3 replaces a purely capability-oriented 1.2 signal with a concrete selection tradeoff: standard pricing is unchanged, while a much cheaper contributor tier permits training on inputs and outputs. Its claimed reduction in turns and filler remains unverified, so the new version is best treated as a workload test whose cost advantage must be weighed against data policy.

Context Map
modelcoding#coding-agents#model-selection
Uncertainty
The discount carries a data tradeoff: Meta may use contributor-tier inputs and outputs for model training. The material provides no benchmark results for the claimed coding improvements, and both tiers retain Muse Spark 1.2 pricing.