Meta released Muse Spark 1.1 on Thursday, a multimodal agentic coding model priced at $1.25 per million input tokens and $4.25 per million output tokens. The pricing puts Muse Spark 1.1 slightly above Anthropic's Claude Haiku 4.5 and OpenAI's GPT-5.6 Luna, the two models it most directly targets. Meta first previewed Spark in April and is now opening the production version to developers as the agentic coding category consolidates around a handful of high-volume, low-cost models.
The pitch is enterprise workload coverage: multistep reasoning, bug fixing, large code migrations, and orchestration across external apps and services. Meta is positioning Muse Spark 1.1 for the same buyer that has been standardizing on Claude Haiku 4.5 for background agents and GPT-5.6 Luna for cheap, high-throughput code tasks. The token pricing is the negotiating table — every cent per million tokens compounds when a single autonomous coding run consumes tens of millions of tokens.
Meta describes the model's strengths in its own launch post.
“Muse Spark 1.1 delivers exceptional performance in personal agentic tasks that require planning and orchestration across a range of external apps and services”— Meta, Company blog post
Key facts
- 01Meta released Muse Spark 1.1 on Thursday, priced at $1.25 per million input tokens and $4.25 per million output tokens.
- 02Pricing lands slightly above Anthropic's Claude Haiku 4.5 and OpenAI's GPT-5.6 Luna, the two most direct competitors.
- 03Mark Zuckerberg posted on X to promote the launch — his first post on the platform since July 2023.
- 04Meta also released Muse Image, an AI image-generation model, earlier the same week on Tuesday.
- 05The launch coincided with OpenAI's GPT-5.6 family debut on the same Thursday.
The company says Muse Spark 1.1 is designed to deploy new features inside enterprise systems and manage digital workflows without human hand-off between steps. That's the same capability envelope Anthropic and OpenAI have been shipping against for more than a year, which puts Meta in the position of catching up on a category it did not define.
The launch was significant enough internally that Mark Zuckerberg broke a three-year silence on X to promote it. His previous post on the platform was in July 2023, roughly when the service rebranded from Twitter. On Thursday he called Muse Spark 1.1 a coding model with strengths in agentic performance, tool use, and computer use, and signaled that more model releases are coming soon.
Zuckerberg framed the value proposition directly.
“a strong agentic and coding model at a very low price”— Mark Zuckerberg, Meta CEO
The Spark 1.1 release caps a dense week for Meta. On Tuesday the company unveiled Muse Image, an AI image-generation model, and on Thursday it pushed out Spark 1.1 alongside a coincident GPT-5.6 launch from OpenAI. The Meta Model API — which we covered earlier this week as the delivery mechanism for Spark 1.1 — is how developers get access to both.
Meta is showing up late to the agentic coding fight. Anthropic has been shipping Claude models tuned for autonomous coding for well over a year, and OpenAI's Codex agent and the GPT-5.6 family cover the same ground. That said, Muse Spark 1.1's pricing is close enough to Claude Haiku 4.5 and GPT-5.6 Luna that enterprise buyers running large volume have a reason to at least benchmark it — and multi-model routing setups make it cheap to try.
The open question is quality-per-dollar on real coding workloads, not marketing benchmarks. Zuckerberg's claim that Spark is strongest at tool use and computer use is a testable one, and enterprise teams that already have evals wired up against Claude and GPT-5.6 will have Spark scores within days. If Muse Spark 1.1 lands within a few percentage points of Haiku 4.5 on agentic reliability, the slightly higher token price becomes a rounding error against the switching cost of moving off an incumbent.
Meta's real leverage in this market is not pricing — it's distribution. The company has a massive developer surface across WhatsApp Business, Instagram, and Messenger, and any of those channels are natural homes for embedded coding agents built on Muse Spark. The three-year Zuckerberg X hiatus ending on a model launch says more about how Meta ranks this bet than any pricing table. Anthropic and OpenAI now have a third well-capitalized competitor pricing into their most lucrative developer segment, and the ceiling on agentic coding margins just moved down again.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
