Muse Spark
TrialPlatforms
A Meta model shipped in April and cited as the predecessor to the internal Watermelon model.
Why it's here
Placed in Trial: 9 article(s) of evidence from 4 source(s), led by model releases, with 4 in the last 30 days. Confidence 70%.
Evidence (9)
- 7Ars Technica AI·8/10/2026model_releaseMeta Reboots AI Strategy Around Open-Weight Models
Meta said it will focus on open-weight large language models and released an open model called Muse Glimmer. The company also said it will open the weights for its stronger model, Muse Spark 1.2, in the coming weeks, while Mark Zuckerberg published a long essay explaining Meta's AI and governance approach.
- 7The New Stack·8/10/2026product_launchMeta Launches Muse Code, Cheaper Than Claude Code but With Data Tradeoffs
Meta released Muse Code, its first AI coding agent, built on Muse Spark 1.2 and priced lower than Anthropic's comparable offering. In testing, the agent handled the same coding tasks as Claude Code, but Meta's default contributor tier can use customer code, prompts, and sessions for product improvement.
- 7Simon Willison·8/5/2026model_releaseMeta releases Muse Code beta and Muse Spark 1.2
Meta announced Muse Code, a beta terminal coding agent designed for complex software engineering work across large codebases. The release is paired with Muse Spark 1.2, a coding-focused model update that emphasizes code generation, debugging, repository understanding, and long-horizon developer workflows.
- 7The New Stack·8/5/2026researchMeta uses engineer code fixes to train internal AI coding tools
Meta is asking engineers to submit code fixes made while using its internal coding agent, MetaCode, so those corrections can help train and post-train future models. The effort has already produced more than 800 fixes from 7,000 weekly active users, and Meta says the data has improved Muse Spark 1.1 and will help train an upcoming model codenamed Watermelon.
- 8The New Stack·7/11/2026model_releaseOpenAI, SpaceXAI and Meta spark an AI price war
OpenAI, SpaceXAI, and Meta each launched or expanded new AI models within a 24-hour window, but the shared headline was pricing rather than capability. The article argues that frontier AI is increasingly competing on lower token costs and efficiency, signaling growing pressure on margins across leading model providers.
- 5The New Stack·7/9/2026model_releaseMuse Spark 1.1 adds API access and improved tool use
Muse Spark 1.1 is the first Spark model to expose an API, following the original release in April. Meta says the model improves agentic tool calling and computer use, and the post notes a new LLM plugin that enables CLI and Python access to the model.
- 3Simon Willison·7/9/2026framework_updatellm-meta-ai 0.1 released
Simon Willison released llm-meta-ai 0.1, a plugin for the LLM tool that adds support for the Meta AI API. The update lets LLM run prompts against Meta's new muse-spark-1.1 model. It is a small developer-tool release focused on expanding model access rather than introducing a new model.
- 7Hacker News·7/9/2026model_releaseMeta releases Muse Spark 1.1
Meta has released Muse Spark 1.1, accompanied by evaluation material and developer documentation. The announcement appears to focus on model capabilities and how developers can build with the Muse Spark system.
- 6The New Stack·7/8/2026researchMeta Claims Its Internal Model Caught Up to OpenAI, but Offers No Benchmarks
Meta AI chief Alexandr Wang reportedly told staff that the internal model codenamed Watermelon has matched OpenAI’s GPT-5.5 on unnamed benchmarks. The claim was presented without benchmark details, model cards, or reproducible evaluation data, leaving it unverified.