Skip to content
NewsModel release

Meta Superintelligence Labs opens its first commercial API for a reasoning model with Muse Spark 1.1

· by Pondero Newsdesk

The short version

Meta Superintelligence Labs launched a public preview of Muse Spark 1.1 on the Meta Model API on July 9, 2026. The multimodal reasoning model carries a 1-million-token context window, costs $1.25 per million input tokens, and accepts both OpenAI and Anthropic SDK integrations.

Meta Superintelligence Labs opens its first commercial API for a reasoning model with Muse Spark 1.1

Meta Superintelligence Labs opened its first commercial API for a reasoning model on July 9, launching Muse Spark 1.1 in public preview at $1.25 per million input tokens and $4.25 per million output tokens, per Meta's developer blog. The API accepts both the OpenAI SDK and the Anthropic SDK. New accounts receive a one-time $20 credit to test the model before committing to pay-as-you-go usage.

What happened

Meta Superintelligence Labs announced Muse Spark 1.1 on July 9, 2026, and opened the Meta Model API to US developers at the same time, per Meta's AI blog. The model is a multimodal reasoning model built for agentic tasks: tool use, computer use, coding, and multimodal understanding. Its context window spans 1 million tokens.

The original Muse Spark was announced in April 2026 with no public developer API. The 1.1 release is the first version available for direct commercial use. The API is US-only during the public preview period.

Pricing is $1.25 per million input tokens and $4.25 per million output tokens, per Meta's developer blog. Reasoning tokens, generated before the model answers, count as output and are billed at the $4.25 rate. Developers can control reasoning depth through the reasoning_effort parameter, which runs from minimal to xhigh.

Both the OpenAI Chat Completions format and the Anthropic Messages format are supported, pointing at the same api.meta.ai/v1 base URL with a Meta Model API key. For teams already using an OpenAI-compatible client, connecting to Muse Spark 1.1 requires changing the base URL, the API key, and the model name to muse-spark-1.1, per the developer blog. OpenCode users can connect through /connect and selecting Meta Model API in terminal.

Additional capabilities per Meta's developer blog: built-in web search grounding (add {"type": "web_search"} as a tool on any Responses API call), multi-agent orchestration as either a main agent or a subagent, repository-level code edits, bug diagnosis in complex codebases, computer use across desktop applications, and multimodal reasoning over images, video, and documents.

TechCrunch characterized the pricing as "in line with, albeit slightly above, Anthropic's Claude Haiku 4.5 and OpenAI's GPT-5.6 Luna" based on Reuters reporting. Muse Spark 1.1 also became available on the same date in "Thinking" mode in the Meta AI app and on meta.ai.

Why it matters

Before July 9, developers who wanted a Meta-hosted reasoning model had two options: run open-weight Llama locally, or use Meta AI consumer products. Neither offered commercial SLA, standard API billing, or direct SDK compatibility. The Meta Model API changes that.

The OpenAI and Anthropic SDK compatibility is the practical accelerant. Most teams running hosted LLM workloads already have client code targeting those two providers. Pointing that code at Muse Spark 1.1 requires a configuration change, not a rewrite. That lowers the cost of evaluation for any team that wants to compare it against their current provider on a specific task.

The 1-million-token context window at this price tier is worth specific attention. Agentic coding workloads over large codebases, and multi-step computer-use sessions, consume context quickly. A longer window at comparable pricing means fewer mid-session resets and less context management overhead. Teams running cost-sensitive agentic workflows now have a third option to benchmark at this tier.

The US-only restriction during public preview limits the audience for now. Teams outside the US will need to wait for regional expansion before they can use the API at all.

Context and reactions

Meta CEO Mark Zuckerberg posted on X about the Muse Spark 1.1 launch on July 9, his first post on the platform since July 2023, per TechCrunch. He called Spark "a strong agentic and coding model at a very low price" and said it was "strongest at agentic performance, tool use, and computer use," and noted "more to come soon."

The launch landed in a crowded week. OpenAI released GPT-5.6 on the same day. SpaceXAI released Grok 4.5 on July 8. TechCrunch noted that Anthropic and OpenAI had offered similar agentic models for some time, placing Meta as a later entrant into this market. Meta also released Muse Image, a separate image-generation model, the same week.

What to watch next

The public preview will move to general availability at some point, likely with a pricing announcement and possible regional expansion beyond the US. Third-party benchmarks comparing Muse Spark 1.1 against Claude Haiku 4.5 and GPT-5.6 Luna on identical agentic coding tasks will show whether the 1-million-token context advantage translates to measurable throughput gains at the same spend. Meta's large Llama developer community represents the most likely early adopter pool; adoption figures there will signal how much traction the commercial API strategy gains.

Sources