Skip to content
Rush Commerce
Software & Dev2 min read

Meta's Muse Spark undercuts coding AI — and closed the door

Meta launched Muse Spark 1.1, an agentic coding model at $1.25/$4.25 per M tokens — API-only, no open weights. The Llama portability story is over. Build behind an abstraction.

Mark Zuckerberg posted on X for the first time in three years to launch a coding model — which tells you how much Meta needs this one to land. Muse Spark 1.1, released July 9, is an agentic coding and computer-use model priced at $1.25 per million input tokens and $4.25 per million output — roughly a quarter of what the flagship OpenAI and Anthropic models cost. The cheaper price is the headline. The quieter story is what Meta didn't ship: open weights.

What actually happened

Per TechCrunch, Muse Spark 1.1 is built for "large agentic workloads" — fixing bugs, implementing features in enterprise systems, and running large code migrations. Meta's own announcement touts a 1-million-token context window, zero-shot use of new tools and MCP servers, and parallel subagent delegation. Zuckerberg called it "a strong agentic and coding model at a very low price."

Here's the pivot. Meta built its reputation on Llama — open-weight models you could download, self-host, and own. Muse Spark is API-only, accessed through the new Meta Model API (an OpenAI-compatible endpoint in public preview). There is no weights download. The company that made "open" its wedge against OpenAI just shipped its flagship coder as a metered, closed API like everyone else.

Why it matters for your business

Two things are true at once. First, the price war on capable coding models is real and it's in your favor — a model at $1.25/$4.25 that can run migrations and fix bugs makes agentic dev work cheaper by the month. Route the boring, high-volume work to the cheap model and you'll feel it on the bill.

Second, don't mistake "cheap" for "portable." If you adopted Llama specifically so you could self-host and never be repriced, Muse Spark is not that — it's another rented endpoint that Meta controls the price and availability of. The move that protects you is the same one it's always been: put every model behind an abstraction you own, so swapping Muse Spark for Claude Haiku, GPT-5.6 Luna, or a self-hosted open model is a one-line change. Meta's OpenAI-compatible API actually makes that easy — take the interoperability, skip the lock-in.

Key takeaways

  • Meta launched Muse Spark 1.1 on July 9 — an agentic coding/computer-use model at $1.25/$4.25 per M tokens, ~1/4 of flagship rivals
  • It's API-only via the new Meta Model API (OpenAI-compatible) — no open weights, unlike Llama
  • The cheap price is genuinely useful; route high-volume agentic dev work to it and watch the bill
  • Don't confuse cheap with portable — keep every model behind an abstraction you own so you can swap on demand

Want the cheap models without the lock-in? We build coding and automation pipelines behind a model layer you own — route to the cheapest capable model today, swap it out tomorrow, no rewrite. See what we build or estimate what owning your model layer is worth.

Sources: TechCrunch, Meta AI.

  • #meta
  • #muse-spark
  • #agentic-coding
  • #model-routing
  • #portability
TR

Tommy Rush — Founder, Rush Commerce

Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More

Get The Rush Report weekly — one email, zero fluff.