Meta Launches Muse Spark 1.1 in US Public Preview With Lower-Cost Pricing for Coding Agents

Meta announced Muse Spark 1.1 is now available in the US public preview, giving developers access to tools that can handle coding assistance, file inspection, and more complex software tasks.The model is priced starting at $1.25 per million input tokens and $4.25 per million output tokens, making it a more affordable option compared to similar models from OpenAI and Anthropic. US developers participating in the preview receive $20 in free credits before any charges apply.Muse Spark 1.1 features a context window of one million tokens, according to Meta.What Muse Spark 1.1 Does and How It’s PricedMeta introduces Muse Spark 1.1, a model designed for various tasks including coding and media understanding, across different input types. It can handle writing and debugging code, using software tools, interpreting text, images, video, and documents, and completing multi-step tasks with minimal human oversight.Developers can ask the model to interpret written instructions, review screenshots, read project files, or operate within software environments as part of a single job. Meta states that the model excels at agent performance, tool use, and computer tasks.Mark Zuckerberg described Muse Spark 1.1 on X as “a strong agentic and coding model at a very low price,” emphasizing its strengths in agent performance and tool utilization.The API pricing for Muse Spark 1.1 is $1.25 per million input tokens and $4.25 per million output tokens.The US public preview offers $20 in free credits for new users, allowing testing without immediate payment. Meta’s chief AI officer, Alexandr Wang, told CNBC that the pricing is “very aggressive and attractive” and that Meta aims for a pricing model that scales with high usage.Compared to other coding-focused models, Muse Spark 1.1’s costs are lower. Anthropic Claude Sonnet 5 charges $2 for input and $10 for output per million tokens, with an introductory rate valid through August 31, 2026, after which prices rise to $3 and $15 respectively.Anthropic Claude Opus 4.8 charges $5 for input and $25 for output per million tokens. OpenAI’s GPT-5.6 pricing has not yet been publicly announced for broad availability.Overall, Muse Spark 1.1 offers a competitive option at both input and output pricing levels.API Compatibility, Free Credits, and How to Access the PreviewMeta has indicated that Muse Spark 1.1 is compatible with OpenAI-style APIs, which could make it easier for teams to compare it with other AI models. This means developers switching providers or testing multiple models might find it simpler to use similar tools and integration approaches, reducing the amount of rework needed. WhileMeta hasn’t provided specific details about the compatibility parameters, the mention of OpenAI-style framing suggests an API design that aligns with patterns established by OpenAI’s Chat Completions API.For developers interested in testing Muse Spark 1.1, Meta offers a US public preview through its developer platform. New preview users can redeem $20 in free credits.To get started, set up API access using the endpoints provided by Meta. This allows testing of coding assistants, app development, and agent-based workflows before deciding to move to paid usage.Meta has not shared details about regional availability beyond the US public preview. Developers outside the US should keep an eye out for announcements regarding international access.What Teams Should Consider Before Production Use and Current AvailabilityFor engineering teams considering Muse Spark 1.1 in their production coding workflows, there are several practical considerations to keep in mind. It is important to establish rules for where the tool can operate and when human approval should be required before any code changes are made.Setting up logging to track changes made by the agent is also advisable, especially for AI-assisted fixes that might cause issues downstream.Defining review requirements for pull requests generated or modified by the agent helps maintain oversight. Additionally, sandboxing agent operations can prevent unintended changes to critical systems.While the lower pricing now removes cost as a barrier to experimentation, it does not reduce the operational risks associated with running coding agents at scale. The responsibility for code quality and system stability remains with the engineering team using the tool.Meta’s launch positions it as a serious competitor in the agentic coding market, alongside companies like Anthropic, OpenAI, and Google. Meta previously introduced Muse Spark as an AI assistant for its smart glasses and apps. Now, Muse Spark 1.1 is aimed at a different audience: developers who want to integrate the model into their coding tools and workflows.This move fits with Meta’s broader push into infrastructure services beyond social media. The company has been expanding its enterprise AI efforts, including investing in custom silicon and data center partnerships to support large-scale model training and inference.Muse Spark 1.1 is currently available in the US through Meta’s developer platform in a public preview. The launch includes a $20 free credit and a preview pricing structure, which are active from Thursday.Meta has not announced a general availability date or a timeline for regional expansion. Developers can keep an eye on Meta’s developer channels for updates on pricing, new features, and availability outside the US.

Source link