2 min read AI-generated

Meta Jumps In: Muse Spark 1.1 Takes On Claude and GPT

Copy article as Markdown

With Muse Spark 1.1, Meta ships its second Superintelligence Labs model — an agentic coder with computer use and a 1M-token context. Zuckerberg broke a three-year X silence to launch it.

Featured image for "Meta Jumps In: Muse Spark 1.1 Takes On Claude and GPT"

Meta has arrived in the coding race. On July 9 the company unveiled Muse Spark 1.1 — the second model out of its Superintelligence Labs and the first serious push into territory Claude, GPT, and Gemini had mostly split between them. Notable enough that Mark Zuckerberg posted on X for the first time in three years to launch it.

What Muse Spark 1.1 does

Muse Spark 1.1 is a multimodal reasoning model built for agentic tasks. The big gains, per Meta, are in tool and computer use, coding, and multimodal understanding. In practice that means: diagnosing and fixing complex bugs, shipping new features into enterprise systems, running large code migrations — the standard promises of this agent generation. Plus a one-million-token context window.

The strategy behind it is the interesting part. Muse Spark is proprietary and closed-weight — unlike Meta’s open Llama family. So Meta is deliberately taking the same road as Anthropic and OpenAI here: the flagship stays closed.

Price and access

Muse Spark 1.1 is free in Thinking mode inside the Meta AI app and at meta.ai — login required. For developers there’s the new Meta Model API, in public preview for US developers since July 9. Pricing: $1.25 per million input tokens, $4.25 per million output tokens, plus $20 in starter credits.

That plants Meta firmly in the affordable tier. For comparison: Claude Fable 5 runs at $10 / $50. So Muse Spark isn’t playing in the same price class as the absolute frontier models — it’s aiming at the volume business, where cost per token is what decides.

My take

The market for coding agents is getting crowded. Within a few days we got Grok 4.5 from xAI, GPT-5.6 from OpenAI, and now Muse Spark 1.1 from Meta — all with the same buzzwords: agentic, long-running, computer use. For us users that’s good news at first: more choice, more price pressure.

Whether Meta actually keeps up remains to be seen — benchmarks are one thing, the real working day is another (see the Cognition case study on Fable 5). But the fact that Meta, long the champion of open models, is now launching a closed flagship says a lot about where the industry thinks the money is right now. Zuckerberg’s return to X was no accident.


Sources: