Frontier Models
Meta Releases Muse Spark 1.1 Agentic Coding Model via Public API
The upgrade positions Meta as a competitor in agentic coding with an 80 percent benchmark score and million-token context, available through a new developer API and consumer app.
Muse Spark 1.1 is a multimodal reasoning model built for agentic tasks with major gains in tool use, computer use, coding, and multimodal understanding.
The launch of Muse Spark 1.1 marks Meta's latest effort to advance agentic capabilities in coding and reasoning models. Developers can access the model through a public preview of the new Meta Model API. The model also appears in Thinking mode within the Meta AI app and on meta.ai.
What background led to Meta developing Muse Spark 1.1?
Meta has positioned the new model as a direct response to offerings from OpenAI and Anthropic in the agentic space. The company aims to provide an affordable alternative through its public API. This approach allows broader access for developers working on coding agents. The development reflects broader industry trends toward agentic systems that can perform complex coding autonomously.
Meta Superintelligence Labs has focused resources on these advancements. Competition in the space has intensified with multiple companies releasing similar tools. Previous iterations in the Muse Spark family laid the groundwork for these improvements. The focus remains on enhancing tool use and computer use functions.
The release comes at a time when agentic coding tools are gaining traction among professional developers. Meta seeks to differentiate through a combination of performance and accessibility. The public nature of the preview invites immediate experimentation from the community.
What details define the Muse Spark 1.1 upgrade?
Muse Spark 1.1 brings major gains in tool use, computer use, coding, and multimodal understanding. The model handles images, video, and PDFs as part of its multimodal support. It includes built-in search with citations for better reasoning. Structured output and parallel tool calling come in an OpenAI-compatible package.
These elements create a complete agentic foundation according to early users. The upgrade builds on prior versions with specific enhancements in coding performance. Public preview availability distinguishes this release from previous internal-only models. Thinking mode integration in the consumer app provides immediate user access.
The combination of API and app availability broadens the user base significantly. Early partners have already begun to integrate the model into their workflows. The focus on coding agents addresses a growing demand in software development.
What are the technical specifics of Muse Spark 1.1?
Muse Spark 1.1 supports a 1 million token context window. The model can delegate execution to parallel sub-agents for complex tasks. Performance improvements appear across benchmarks in various domains per the evaluation report. The model excels in frontend and design coding tasks in particular.
Multimodal understanding extends to video and PDF processing. Reasoning capabilities have been strengthened in the latest version. Tool use improvements allow the model to interact with external systems effectively. Computer use features enable screen and interface interactions in agentic workflows.
Coding abilities cover a range of languages and frameworks with high accuracy. The 1 million token context supports analysis of large projects without truncation. Parallel tool calling increases efficiency in multi-step processes. The overall architecture prioritizes agentic functionality.
| Feature | Specification | Notes |
|---|---|---|
| Context Window | 1 million tokens | Enables processing of extensive documents and codebases |
| Benchmark Performance | 80.0 percent | On Terminal-Bench 2.1 for agentic coding |
| Agent Capabilities | Parallel sub-agents | Allows delegation of execution tasks |
| Multimodal Support | Images, video, PDFs | Includes built-in search with citations |
How does the release affect the market and stakeholders?
Early partners include Replit, Cline, and Box. These companies praise the agentic foundation and enterprise capabilities of the model. The public API opens access to a wider range of developers and organizations. Enterprise users gain from the structured output and tool calling features.
The release may pressure other providers to adjust their pricing or access models. Integration with existing platforms like Replit could accelerate adoption. Replit has highlighted the model's suitability for their platform. Cline and Box have noted the enterprise-grade features in their feedback.
The affordable aspect may attract startups and smaller teams previously priced out of similar tools. Stakeholders in the AI coding space will likely evaluate integration options. Market dynamics could shift as more accessible models enter the arena. The emphasis on enterprise capabilities opens doors in corporate environments.
- Access the public preview through the Meta Model API to begin integration.
- Test the model in Thinking mode within the Meta AI app for consumer applications.
- Review partner feedback from Replit, Cline, and Box for use case ideas.
- Consult the evaluation report for detailed benchmark comparisons.
- Monitor announcements from Meta Superintelligence Labs for future updates.
What reactions have experts offered on Muse Spark 1.1?
Alexandr Wang highlighted the model on social media as the strongest yet for agentic and coding work. The announcement emphasized availability in the Meta Model API. Industry observers note the OpenAI-compatible design as a practical advantage. The reaction aligns with the technical claims in the official announcement.
The comprehensive quote from Replit leadership points to the model's all-in-one design. This feedback suggests strong potential for real-world agentic applications. Endorsements from industry leaders validate the technical advancements made by the Meta team.
What’s most impressive about Muse Spark is how much it packs into one model: massive million-token context, full multimodal support (images, video, PDFs), built-in search with citations, strong reasoning, top-tier coding abilities (particularly frontend and design), structured output, and parallel tool calling — all in a clean OpenAI-compatible package. A complete agentic foundation.Amjad Masad, CEO of Replit
What comes next in Meta's model development roadmap?
Meta plans to continue iterating on the Muse Spark family of models. Further improvements in agentic tasks are expected based on the current trajectory. The evaluation report indicates ongoing work across multiple domains. Developers should monitor the Meta Model API for updates.
Consumer access through the Meta AI app may expand with additional modes. Future releases may build on the evaluation report findings. Expanded context windows or additional modalities could appear in subsequent versions. The API may see more partners announced over time.
Users can expect continued focus on agentic tasks from the Meta team. The roadmap likely includes refinements based on developer feedback. The company will likely release additional evaluation data as new versions emerge.
Frequently asked
What is the context window size for Muse Spark 1.1?
Muse Spark 1.1 supports a 1 million token context window.
Which companies have partnered early with Muse Spark 1.1?
Early partners include Replit, Cline, and Box.
How can developers access Muse Spark 1.1?
Muse Spark 1.1 is available in public preview via the Meta Model API and in Thinking mode in the Meta AI app.