Urgent.News

What's breaking now, across thousands of outlets.

AI

Zuck's Muse to Spark joy with open weights release 'soon'

While you wait, Meta says it’s taught the model to stop wasting tokens and ask for help a bit more often

Zuck's Muse to Spark joy with open weights release 'soon'

Meta, led by CEO Mark Zuckerberg, will soon introduce an open weights version of its flagship AI model Muse Spark. In the interim, Muse Spark 1.3, a smarter, more efficient iteration of the model, is now available on Facebook's API service and Muse Code CLI. Since the initial release of Muse Spark in April, Meta has rolled out multiple refinements to maintain its competitive edge and convince investors that its significant capital expenditure is justified.

Muse Spark 1.3 brings several notable enhancements, particularly for users employing the model with AI agents and code assistants. Independent benchmarking by Artificial Analysis corroborates Meta's claims, demonstrating a 4-point increase in overall intelligence, placing Muse Spark in contention with leading AI models such as GPT 5.6 Sol, Claude Opus 5, and Grok 4.6 High. However, a crucial reasoning mode used for benchmarking will only be accessible post-security review.

Priced at $1.25 for a million input tokens, $0.15 for cached input, and $4.25 for output, Muse Spark 1.3 holds up better in long-horizon workloads and is more realistic about its capabilities. The model requests human counsel more frequently when faced with ambiguous prompts, solicits assistance from the user when stuck, and confirms before executing consequential actions. These enhancements also reduce the number of turns and tokens required to complete a task, potentially lowering costs for users.

For those seeking more affordable options, Meta offers a contributor tier with rates as low as $0.002 for cached input, $0.10 for input, and $0.20 for output per million tokens.

Written by urgent.news from The Register's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Also reported by 1 other outlet

Read the original at theregister.com →

More in AI

More from Wednesday 2 September →