Meta Launches Paid Muse Spark 1.1 AI Model to Challenge Competitors
Meta Platforms released Muse Spark 1.1 today, an upgraded artificial intelligence model designed for computer coding and automated agent workflows.
The release accompanies Meta's first-ever paid application programming interface (API) for software developers. This launch places Meta in direct competition with Anthropic and OpenAI for commercial software engineering tools.
The original version of Muse Spark debuted in April with noted coding limitations. The 1.0 version scored 59.0 on public coding benchmarks compared to 80.8 for Anthropic's Claude and 75.1 for OpenAI's GPT-5.4. Meta built the 1.1 upgrade specifically to close this performance gap.
Meta Chief AI Officer Alexandr Wang announced the model capabilities on social media.
"Today we're releasing Muse Spark 1.1 -- a strong agentic and coding model at a very low price. It's available through our new Meta Model API and in Meta AI."
Wang stated that the system completes long-running tasks across desktop, mobile, and browser interfaces by delegating work.
"Muse Spark 1.1 is strongest at agentic performance, tool use, and computer use. It does well on long-running tasks with 1M token context window, can delegate execution to sub-agents running in parallel, and is trained to use computer interfaces on desktop, mobile, or browser."
The new model features a one-million-token context window that processes entire codebases in a single session. The software operates multi-agent systems by acting as a main coordinator that splits large tasks among parallel sub-agents. It automates actions across desktop applications by writing custom automation scripts or interacting with visual software interfaces directly.
The public preview of the Meta Model API is available for developers based in the United States. Meta prices the service at $1.25 per million input tokens and $4.25 per million output tokens. New users receive $20 in free trial credits before moving to pay-as-you-go commercial billing.
This pricing structure undercuts Anthropic's Claude Sonnet 4.6 model. The rates remain higher than budget tiers like OpenAI's GPT-5 mini and Anthropic's Claude Haiku 4.5.
Early enterprise partners include software companies Replit, Cline, and Box. Replit Chief Executive Officer Amjad Masad praised the processing capacity of the new model.
"Meta is clearly building for serious agentic coding – strong tool use at a price point that makes it viable to run real coding workloads at scale."
Cline Chief Executive Officer Saoud Rizwan noted the price benefits for large operations.
"That combination is rare, and it's exactly why we wanted Cline developers to have access early."
Meta plans to deploy the model immediately into its main consumer products. Muse Spark 1.1 will operate in "Thinking" mode on the Meta AI web platform and mobile app. The company will use the model to replace previous Llama models that power automated assistants inside Facebook, Instagram, WhatsApp, and Ray-Ban smart glasses.
The deployment represents a shift away from Meta's previous strategy of distributing open-source AI models for free. Chief Executive Officer Mark Zuckerberg and Chief AI Officer Alexandr Wang shifted the division toward closed-source models to generate direct monetization.
Before the public release, Meta Superintelligence Labs evaluated the safety profile of the model under its Advanced AI Scaling Framework. Initial tests without safeguards met high-risk thresholds for potential misuse in cybersecurity and biochemical domains. The team implemented multi-layered engineering mitigations that reduced the final residual risks to a moderate or lower rating.
