Meta Releases Muse Spark 1.1 and Opens the Meta Model API - A Full Entry into the AI Coding Market

Meta announced Muse Spark 1.1, a multimodal reasoning model built for agentic tasks, on July 9, 2026¹. Alongside it, the company opened the Meta Model API - a way for developers to build with the model - in public preview¹. It is the latest in a string of releases from the Superintelligence Labs organization, following this week’s Muse Image and Muse Video, and TechCrunch frames it as Meta’s entry into an AI coding market where Anthropic and OpenAI got there first².

The original Muse Spark was announced in April 2026 but was only available to “select partners”². With this release, ordinary developers can use the model through an API for the first time, and it also went live the same day in “Thinking” mode in the Meta AI app and on meta.ai¹.

A Model Designed Around Agentic Coding

Muse Spark 1.1 claims major gains in tool use, computer use, coding, and multimodal understanding¹. For coding, it targets diagnosing and fixing complex bugs, implementing new features in enterprise-grade systems, and executing large code migrations¹, and it is positioned as strong at computer-use workflows that unfold across multiple applications¹.

A distinctive aspect is that the model’s role as an AI agent is explicitly designed in. As a main agent, it can gather context, make a plan, and delegate execution across parallel subagents. Running as a subagent, it sticks to its assigned job and knows when to escalate back to the main agent¹. The context window is 1 million tokens, and the model actively manages how it is used¹.

On safety, Meta says the model shows strong resistance to direct jailbreaks, prompt injection, and developer-prompt attacks, along with lower hallucination rates and reduced sycophancy¹.

An OpenAI-Compatible API at a Low Price Point

The Meta Model API is delivered as an “OpenAI-compatible” package and supports features common in agentic coding tools, such as planning mode, goal conditioning, subagent delegation, and context compaction¹. The apparent aim is to lower the barrier for developers to switch from - or run alongside - existing OpenAI-based code and tooling.

On pricing, TechCrunch reports $1.25 per million input tokens and $4.25 per million output tokens, with $20 in free credits for every new API account². The publication places this price point in competition with Anthropic’s Claude Haiku 4.5 and OpenAI’s GPT-5.6 Luna². CEO Mark Zuckerberg described it as “a strong agentic and coding model at a very low price,” citing agentic performance, tool use, and computer use as its strengths².

A Turning Point Away from the Open-Model Playbook

Meta has so far centered its AI strategy on distributing models, most notably the Llama series. Muse Spark 1.1 is the company’s first full-scale move into metered, paid API access. Meta itself calls the model “a significant upgrade from Muse Spark” that is “competitive with leading alternatives”¹, and says the release - together with this week’s Muse Image launch - brings the company closer to its “personal superintelligence” vision¹.

The timing is telling. A day earlier, OpenAI made GPT-5.6 generally available with its low-cost Luna tier at $1/$6, and Grok 4.5 also entered the market on a low-price, high-efficiency pitch. Alongside the frontier performance race, competition over models that can run everyday coding and agent workloads cheaply is heating up fast - which, for anyone choosing models to build into their tools, means more options on both price and compatibility.