Muse Spark 1.1
Points and comments are a snapshot, not live.
Meta releases Muse Spark 1.1, a highly capable multimodal reasoning model for agentic tasks.
Muse Spark 1.1 shows major gains in tool use, computer use, coding, and multimodal understanding. It features a 1 million token context window and can orchestrate multi-agent systems. The model is available via a new Meta Model API and in Meta AI's Thinking mode. Pricing is $1.25 per million input tokens, $4.50 per million output, and $0.15 for cached input. Meta says it excels at complex projects faster than its predecessor by delegating to subagents. It can manage context by compacting earlier work and retrieving relevant information.
What commenters are saying
Commenters focus heavily on pricing. The model is cheaper than Grok 4.5 and Claude Opus/Fable, with unusually low cached input costs ($0.15/Mtok). Some call it "directionally opus level intelligence for haiku prices," which is a big deal for application developers. Others note it's still expensive compared to DeepSeek V4 ($0.004/Mtok cache hits). Several point out that Meta is not currently on most model selectors' radar. The thread broadly welcomes more competition, though some worry the cost advantage may be temporary subsidization to gain users.