Meta Launches Multimodal AI Model Muse Spark Across Social Platforms and Smart Glasses

Meta is pushing generative AI beyond chatbots toward agents capable of planning, calling tools and completing multi-step tasks. Muse Spark 1.3 is positioned as the company’s most capable model, with improvements in coding, long-running multitasking and proactive clarification. Its planned integration into core products including Instagram also makes the release an important part of Meta’s effort to bring agentic AI to its consumer platforms.

Meta recently released Muse Spark 1.3 alongside Muse Glimmer, an open-source model with roughly 30 billion parameters designed for agentic workloads on personal computers. The company says Spark 1.3 uses about 20% fewer tool calls and 25% fewer tokens than Muse Spark 1.2 on agentic coding tasks. Glimmer is intended to run multi-step reasoning and automated workflows locally, offering developers a lighter alternative to Meta’s flagship model.

Meta Platforms built a large developer following by releasing the weights of its Llama models, but its decision to keep the flagship Muse Spark system closed raised questions about that strategy. Muse Glimmer marks a renewed push into open-weight AI, targeting agents that can execute code, call tools and complete multi-step workflows on local machines. Local deployment could reduce cloud costs while allowing companies and individuals to process sensitive information offline.

Meta released Muse Glimmer on Aug. 10, 2026, under the permissive Apache 2.0 license. The 30-billion-parameter model supports multimodal and multilingual reasoning, while 4-bit versions require less than 20GB of memory and can run on a single consumer GPU. Meta said Glimmer outperformed similarly sized Google Gemma 4 31B and Alibaba-backed Qwen3.6 27B models across selected coding, software-engineering and multi-step agent benchmarks, though the results remain company-reported.

Meta Superintelligence Labs introduced the Muse Spark family in April as a multimodal reasoning system designed for complex scientific and software tasks. Muse Spark 1.2 extends that push as leading AI developers increasingly compete not only on headline accuracy, but also on the cost and speed of deploying models. Its tool-free Olympiad tests are significant because they seek to isolate the model’s underlying reasoning rather than gains from web search, coding or calculators.

Meta released Muse Spark 1.2 on Aug. 5 and on Aug. 7 said it achieved gold-medal-level results across several STEM Olympiads without external tools. Vals AI scored the model at 71.9 on the Vals Index, up 3.5 points from Muse Spark 1.1, placing it in the top five at a cost of $0.69 per test. It also led Finance Agent v2 with 59.4%, ahead of Gemini 3.6 Flash at 58.1%, while completing the Index run in 629.7 seconds.

Meta, previously known for its open-source Llama models, is shifting toward a closed-source strategy to challenge OpenAI and Anthropic. Meta Superintelligence Labs, led by former Scale AI executive Alexandr Wang, has introduced Muse Spark, a model focused on agentic tasks and coding. Its low pricing is designed to attract small and midsize development teams, formally marking Meta’s entry into the closed-source frontier-model market.

Meta formally unveiled the upgraded Muse Spark 1.1 on July 9, 2026, expanding its context window to 1 million tokens and opening paid access to the Model API for the first time. Mark Zuckerberg took the unusual step of promoting it in a post. Input and output tokens are priced at just $1.25 and $4.25 per million tokens, respectively, as Meta seeks to undercut rivals and gain market share.

Meta has integrated generative AI search into Facebook, using its Muse Spark model to synthesize information from public posts, Groups and Reels and provide users with consolidated answers. Because the data may also be used for model training and ad personalization, an EU privacy advocacy group alleges that the practice violates the purpose-limitation principle under the General Data Protection Regulation (GDPR). The case could determine whether platforms may repurpose personal data collected for existing uses.

As of July 20, 2026, Meta has formally launched Facebook AI Search, which compiles public information from across the platform and generates summaries. The reliability of its content, source attribution and handling of personal data remain under scrutiny. A complaint has been filed with the EU regulatory system, but no penalty, amount at issue or enforcement timetable has been announced. Meta must also explain the legal basis for using data in model training and ad personalization.