Meta Muse Spark 1.1 Review: 1M Tokens & Multi-Agent AI

Meta rolled out Muse Spark 1.1 on July 9 through the Meta Model API and inside the Meta AI app, and it deserves more attention than its launch received. This isn't a chatbot upgrade — it's Meta's most direct move into the agentic-coding and computer-use space Anthropic and OpenAI have been competing in all year, priced well below both.

What Muse Spark 1.1 Actually Does

Muse Spark 1.1 is a multimodal reasoning model built for agentic work: planning multi-step tasks, calling tools, and operating computer interfaces across desktop, mobile, and browser — not just answering questions about them. It can act as a primary agent that breaks a task into a plan and hands pieces of that plan off to subagents working in parallel, then reassembles the results. That design is meaningfully different from a single model chewing through one long prompt.

The headline spec is the context window: 1 million tokens, and Meta says the model actively manages it rather than just holding it passively. It remembers earlier actions, retrieves relevant information from much earlier in a session, and compacts what it no longer needs — while keeping the steps that matter for whatever comes next. For anyone who has watched a long agent run drift or forget its own earlier decisions, that's the practical problem this is aimed at solving.

The Pricing Angle

Meta priced the API at $1.25 per million input tokens and $4.25 per million output tokens. That undercuts both OpenAI's and Anthropic's flagship-tier pricing by a wide margin, and it lands right in the middle of the industry's broader move toward cheaper, faster models this year. Meta isn't trying to win on raw benchmark supremacy here — it's trying to be the model teams reach for when they're running agents at volume and token cost actually shows up on the bill.

Why It's Worth Trying

If you're building or running agent workflows — coding agents, research agents, anything that has to hold context across a long multi-step task — Muse Spark 1.1 is a legitimate option to benchmark against whatever you're using now, especially if cost per run has been a limiting factor. The combination of a 1M-token context window, native multi-agent orchestration, and computer-use support across three surfaces (desktop, mobile, browser) at this price point is a genuinely new option, not an incremental refresh.

Bottom line: Muse Spark 1.1 is a purpose-built agentic model that makes it easier to run long, multi-step agent workflows without context loss or runaway token costs — and it undercuts the two biggest competitors on price.

It's available now in Thinking mode through the Meta AI app, meta.ai, and the Meta Model API.