Meta Unveils Muse Spark 1.1, an Agentic Coding Model — Zuckerberg Posts on X for the First Time in Three Years

Meta Superintelligence Labs released Muse Spark 1.1, a multimodal model built for agentic coding and tool use, alongside a public preview of the Meta Model API. Priced aggressively at $1.25 per million input tokens, it takes direct aim at OpenAI and Anthropic in the agentic coding market.

TL;DR — On July 9, 2026, Meta released Muse Spark 1.1, a multimodal reasoning model built for agentic coding, tool use, and computer use. Alongside it, Meta opened a public preview of the Meta Model API, putting the company into the same paid-API business as OpenAI and Anthropic. Pricing is set at $1.25 per million input tokens and $4.25 per million output tokens. Mark Zuckerberg posted about the launch on X — his first post there since 2023.

Meta Superintelligence Labs (MSL) announced Muse Spark 1.1 on its official blog on July 9. It follows the original Muse Spark launched in April, and the company says it delivers major gains in agentic tasks that require planning and orchestration, as well as in fixing bugs, implementing features, and running large-scale code migrations across complex codebases.

What Changed — Agents, Coding, and Computer Use

According to Meta, Muse Spark 1.1 actively manages a 1-million-token context window, remembering earlier actions and compacting information while preserving the steps needed for later work. As the main agent, it gathers context, forms a plan, and delegates execution to parallel subagents; as a subagent, it stays within its assigned scope and escalates back to the main agent when needed.

For computer-use workflows, Meta trained the model to choose between writing automation scripts and clicking directly, depending on which is faster in a given situation, rather than reasoning through every step one click at a time. On the coding side, the model was built to adapt smoothly to popular agentic coding harnesses, supporting planning mode, goal conditioning, subagent delegation, and context compaction.

Release date July 9, 2026 (US time)
Model focus Multimodal reasoning, agentic coding, computer use
Context window 1 million tokens
API pricing $1.25 per million input tokens / $4.25 per million output tokens
Availability Meta AI app and meta.ai "Thinking" mode + Meta Model API public preview

Pricing Puts It Near Claude Haiku and GPT-5.6

TechCrunch, citing Reuters, reported that the pricing lands close to (slightly above) Anthropic's Claude Haiku 4.5 and OpenAI's GPT-5.6 Luna. Meta's pitch is straightforward: Spark can handle the kind of large agentic workloads, bug fixes, and large-scale code migrations that enterprises are increasingly turning to AI companies to provide.

Category	Detail
Early partner feedback	Replit CEO Amjad Masad called it "a complete agentic foundation"
	Cline CEO Saoud Rizwan praised "strong tool use at a price point that makes it viable to run real coding workloads at scale"
	Box VP of AI Products Yashodha Bhavnani said it delivered "enterprise capabilities competitive with today's leading frontier models" on Box's internal evaluation set
Safety	Under Meta's Advanced AI Scaling Framework, the model operates within safe margins across Chemical & Biological, Cybersecurity, and Loss of Control risk categories

Meta also launched its first image-generation model, Muse Image, earlier the same week on Tuesday, July 7 — making this a pivotal week for the company's AI product expansion.

Enterprises evaluating coding agents should look beyond benchmark scores alone and compare real-workload token pricing, context management, and compatibility with existing agentic harnesses. Muse Spark 1.1's entry is likely to intensify price competition across the coding-AI market.

Zuckerberg's First X Post in Three Years

Per TechCrunch, Mark Zuckerberg posted about the launch directly on X — his first post on the platform since it rebranded from Twitter in 2023. He described Spark as "a strong agentic and coding model at a very low price," calling it "strongest at agentic performance, tool use, and computer use," and added that there's "more to come soon," hinting at further model releases.

Meta has now shipped the original Muse Spark in April, Muse Image on July 7, and Muse Spark 1.1 on July 9 — a notably faster release cadence since Superintelligence Labs was formed. Whether this pace translates into real enterprise adoption is the next thing to watch.

What It Means

Meta is arriving somewhat late; OpenAI and Anthropic have offered comparable agentic coding models for a while. But the aggressive pricing, million-token context, and compatibility with existing developer tooling could still shift the competitive landscape. The fight over coding AI is increasingly being fought not just on raw capability, but on price and ecosystem compatibility.

Meta Superintelligence Labs released the agentic coding model Muse Spark 1.1 on July 9
Public preview of Meta Model API launched simultaneously — $1.25/$4.25 per million input/output tokens
1-million-token context, subagent delegation, computer use, and coding-harness compatibility all improved
Pricing lands near Claude Haiku 4.5 and GPT-5.6 Luna, challenging OpenAI and Anthropic
Zuckerberg posted about the launch on X — his first post there since 2023