Sakhanda Wire
NVDA $153.64 -1.60% MSFT $506.98 -2.13% GOOGL $279.95 -1.33% META $620.09 -1.01% AMZN $256.57 -0.46%
← Back to the news

Meta Releases Muse Spark 1.3 for Long-Horizon Coding Tasks

Meta Superintelligence Labs has released Muse Spark 1.3, the fourth version of its Muse Spark model in five months. The model targets long-running agentic and coding tasks, supporting multiple workflows within one conversation and adapting as users provide instructions or interruptions. It is available immediately through Muse Code and the Meta Model API, but its weights are closed. The maximum reasoning mode remains restricted pending additional safety testing.

According to Meta, the model asks clarifying questions for ambiguous requests, seeks user input when progress stalls, and confirms consequential actions. It also reports improved ability to recognize its limitations. In internal comparisons with Muse Spark 1.2, Meta engineers measured about 20% fewer tool calls and 25% fewer tokens, alongside less verbosity and cleaner code. The model has a 1M-token context window.

Meta’s published results include 75.4 on DeepSWE v1.1 and 98.5 in the 256K–512K MRCR v2 test. However, the launch scorecard relies on the max mode, which developers cannot currently access. Pricing remains $1.25/M input and $4.25/M output, with a $0.10/$0.20 contributor tier.

Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.

Originally published by MarkTechPost on

Read the original on MarkTechPost ↗

The reporting and the image belong to MarkTechPost and are credited and linked above; the text on this page is Sakhanda Wire's own summary.

← Back to the news

More stories

All the latest news