Meta has officially entered the competitive AI coding sector with the Thursday launch of Muse Spark 1.1, a proprietary large language model designed to optimize agentic workflows and software development tasks. Spearheaded by Meta Superintelligence Labs under AI chief Alexandr Wang, the release marks a strategic pivot from the company’s previously open-source Llama series toward commercially monetized, high-performance models. The new model is now accessible via a public API preview through Meta’s developer portal, with early access extended to select partners and new users placed on a waitlist. Priced to aggressively undercut rivals, Muse Spark 1.1 offers $20 in complimentary credits per new account, followed by rates of $1.25 per million input tokens and $4.25 per million output tokens. Wang emphasized that the pricing structure is engineered to scale efficiently with high-volume enterprise consumption while directly challenging the cost models of competitors like OpenAI and Anthropic. Technical capabilities are optimized for integration with major developer harnesses, including support for emerging agentic frameworks that enable autonomous multi-step task execution. Wang noted that robust coding proficiency remains foundational to building reliable, autonomous AI agents capable of functioning as digital workforces. This launch follows Tuesday’s debut of Muse Image, originally developed under the internal codename Mango, signaling Meta’s broader push to consolidate AI tooling across text, code, and visual generation. The accelerated rollout comes amid mounting investor scrutiny over Meta’s substantial capital expenditures in AI infrastructure, particularly as the company lacks a traditional cloud platform and trails peer labs in deploying widely adopted applications. To address this, Meta is leveraging its proprietary data center infrastructure to host and scale Muse models exclusively on its own platforms, deliberately restricting third-party marketplace distribution for now. Despite the commercial focus, MSL reaffirmed its ongoing commitment to open-source development, confirming that an open-weights variant of Muse Spark is in active training alongside a more powerful successor codenamed Watermelon. Wang declined to disclose release timelines for either project. Industry analysts view Meta’s move as a calculated effort to capture developer mindshare during the rapid expansion of AI agents, a sector that gained significant traction in early 2026 following the mainstream adoption of tools like OpenClaw. By aligning its proprietary models with industry-standard development environments while maintaining competitive pricing, Meta aims to accelerate enterprise adoption and demonstrate measurable returns on its AI infrastructure investments.