Breaking

Meta Releases Muse Spark 1.3 With Gains in Coding and Agentic Work

September 2, 2026 at 16:04 EDT

Foundation Models
AI Agents
Software Dev & Coding

Meta released Muse Spark 1.3 on September 2, saying the update improves coding and agentic performance while making long-running workflows more practical. The model is rolling out in Muse Code and the Meta Model API, with a higher max reasoning mode planned after additional safety testing. Meta’s launch announcement

September 2, 2026 · Meta

Muse Spark 1.3 targets practical agentic work

Meta says the update improves long-running coding and agentic workflows while reducing unnecessary tool use and token consumption.

~20%

fewer tool calls

~25%

fewer tokens

1.3

available in Muse Code and Meta Model API

Where the overhead falls

Meta’s internal comparison reports lower overhead for both tool use and generated output.

~20%

Tool calls cut

~25%

Token use cut

Lower overhead is intended to make extended agentic work easier to sustain.

The key limitation is comparability

Meta describes Muse Spark 1.3 as competitive with frontier models, but a universal coding winner has not been established.

AVAILABLE TODAY

Existing reasoning modes in Muse Code and Meta Model API

COMING AFTER TESTING

Max reasoning, following additional safety testing

Practical gains

• Better support for long tasks and interruptions

• Clarification and confirmation during consequential work

Still to verify

• Independent coding results for the GPT-5.6 Sol comparison

• Whether earlier user complaints persist in 1.3

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.