Meta: 'Watermelon' Model Development on Track
Bubble signal: 0 · NeutralRelevance 35ModelsUS
Source: 36Kr · Published 2026-09-03
Summary
Meta's Chief AI Officer Alexandr Wang said the newly released Muse Spark 1.3 will be priced the same as Muse Spark 1.2 but is more efficient, requiring 25% fewer tokens to run. Meta has not decided whether to open-source the weights of 1.3, but still plans to release 1.2's weights. The company continues development of the larger 'Watermelon' model on schedule, believing it will be highly competitive.
Bubble analysis
This news is mainly about a model release and efficiency improvement, with no specific investment amounts or valuations, so it is weakly related to the AI bubble question. It may shift expectations about AI model competitiveness, but it is several steps removed from the investment cycle.
#meta#model#efficiency
Read original ↗Related signals
linked by shared tags
- OpenAI's new agent is a shot at Meta — but can it compete with free?#meta
- AI Squeezes Its Own Compute: Kimi K3 Writes Its Own 'Super Operators', Inference 3x Faster#efficiency
- Why is Meta stock down on Wednesday?#meta
- Meta Could Be the AI Trade's Come-From-Behind Winner#meta
- Meta Launches AI Travel Agent Muse, Truist Warns of Greater Disintermediation Risk for OTAs#meta