By ByteBot
Meta introduced Muse Spark on April 8 and told developers the API was coming “soon.” It is now June 5 — nearly two months later — and Meta has not published a launch date. Developers who built product plans around Muse Spark cannot call it programmatically, cannot benchmark it independently, and cannot ship anything that depends on it. There is no pricing page, no endpoint documentation, no context window specification. Just a spokesperson statement that Meta “expects to release it this month.”
What Happened
The original plan was to release the Muse Spark API shortly after the April launch. Chief AI Officer Alexandr Wang said it was coming “soon.” Then bugs and infrastructure gaps surfaced in testing, and the timeline slipped to May. May passed without a release. By June 3, the Wall Street Journal and Reuters were reporting that Meta had repeatedly delayed the release and had no scheduled launch date. A Meta spokesperson confirmed the company was testing with select early partners and expected to ship this month — the same language used for May.
Without the API, third-party developers cannot integrate Muse Spark into their products, cannot call it at scale, and cannot run independent evaluations. Artificial Analysis confirmed zero API providers were benchmarking it as of early May. No one outside Meta’s hand-picked partners has production access.
Why This Is More Than a Technical Delay
Meta is launching its first commercial AI developer API. OpenAI, Anthropic, and Google made programmatic access a day-one feature on their model launches — that is the 2026 baseline expectation. Meta is starting from zero on commercial API infrastructure, and the delays suggest the company underestimated what it takes to run a production API at scale. Bugs happen. Infrastructure gaps happen. What does not happen at OpenAI or Anthropic is a two-month gap between model launch and any developer access, with no firm date.
There is also a second trust problem layered under the first. Meta built significant developer goodwill through the Llama family — open weights, self-hosting, fine-tuning. Muse Spark is a proprietary move away from Llama: cloud-only, no downloadable weights, no self-hosting, no migration path. Developers who built on Llama did not get a heads-up that their open-source runway was ending. Now those same developers are waiting on a closed API that has no public timeline. That is two strikes before the first pitch.
What to Use Instead
If your roadmap requires a capable multimodal reasoning API today, here are the options that actually have one:
- Claude Opus 4.7 — Best for coding and agentic workflows. SWE-bench Verified: 87.6%. Pricing: $5/$25 per million tokens. API available now, full documentation.
- Gemini 3.1 Pro — Cheapest frontier option at $2/$12 per million tokens. Native multimodal architecture, accepts video input, strong on vision tasks. API available now.
- GPT-5.4 — Most battle-tested at scale. OSWorld: 75%, above human expert baseline. Best for complex agentic task completion. API available now.
- Open weights — Gemma 4 (Apache 2.0, runs on a 16GB laptop) and NVIDIA Nemotron 3 Ultra 550B are the strongest self-hostable options if you need to stay off proprietary clouds.
For developers waiting specifically for Muse Spark’s health benchmarks — it does lead there, with a HealthBench Hard score of 42.8 versus GPT-5.4’s 40.1 — the alternatives available now are close enough for most use cases. A benchmark advantage does not matter if you are waiting indefinitely.
What to Watch
Meta says it will release the Meta Muse Spark API in June 2026. If that happens, watch for pricing (whether it competes with Gemini 3.1 Pro’s $2/$12), rate limits, context window specifications, and whether early partners have encountered additional issues. A clean launch will go a long way. A third delay would be a serious problem for Meta’s developer ecosystem strategy — and at that point, “soon” stops meaning anything.
The model itself appears strong. But strong models behind closed doors do not build developer ecosystems. When the API eventually ships, Meta will need to earn back the trust it has spent two months burning.
