Meta Launches Paid AI Developer Model "Muse Spark 1.1"

Meta has officially launched Muse Spark 1.1, its first paid AI developer model, marking a significant departure from its open-source Llama strategy.

The model is now available through the new Meta Model API, with pricing set at $1.25 per million input tokens and $4.25 per million output tokens.

This positions Muse Spark above OpenAI's entry-level GPT-5 mini and Anthropic's low-cost Claude Haiku 4.5 but below Anthropic's higher-end Claude Sonnet 4.6. Zuckerberg described the pricing as "very aggressive and attractive," aiming to undercut competitors while scaling usage consumption.

The company is also offering $20 in free credits to developers who sign up for the API. This release represents a strategic pivot for Meta, which previously offered its Llama family of open-weight models for free.

The Muse Spark 1.1 rollout reflects CEO Mark Zuckerberg's broader vision of delivering "personal superintelligence" — AI that can help users pursue their goals, create what they imagine, and take action on what they value most.

The model is expected to replace existing Llama models powering chatbots on WhatsApp, Instagram, Facebook, and Meta's smart glasses.

Muse Spark 1.1 is designed as a multimodal reasoning model built for agentic tasks, with major gains in tool and computer use, coding, and multimodal understanding.

It features a one-million-token context window and can function within multi-agent systems — operating as both a primary agent coordinating tasks and a subagent completing specialized assignments.

The model's key strengths include:

• Agentic performance – Capable of orchestrating complex multi-step workflows with minimal human intervention, including tool use and computer interface navigation.
• Coding capabilities – Diagnoses and fixes complex bugs, implements new features in enterprise-grade systems, and executes large code migrations [citation:1]
• Multimodal understanding – Processes text, images, and video simultaneously, with strengths in visual-to-code artifact generation and ultra-descriptive captioning.
• Computer use features – Maintains context across extended sessions, adapts to evolving requirements, and navigates unfamiliar interfaces [citation:1]
• Competitive performance – Internal testing shows Muse Spark 1.1 outperformed its predecessor and is "competitive with leading alternatives" on Meta's Internal Coding Bench.

"What's most impressive about Muse Spark is how much it packs into one model: massive million-token context, full multimodal support, built-in search with citations, strong reasoning, top-tier coding abilities — a complete agentic foundation." – Amjad Masad, CEO of Replit.

Developer partners, including Replit, Cline, and Box, praised the model's combination of long-context handling, coding capabilities, and attractive pricing for high-volume workloads.

As Zuckerberg noted, Meta's API pricing strategy aims to make advanced AI more accessible to a broader range of developers, potentially reshaping the economics of enterprise AI adoption.