Alisa Davidson
Printed: August 10, 2026 at 9:24 am Up to date: August 10, 2026 at 9:24 am
Edited and fact-checked:
August 10, 2026 at 9:24 am
In Transient
Meta releases Muse Glimmer, a 30B open-weight agentic AI mannequin beneath Apache 2.0 optimized to run regionally on shopper GPUs with 24GB VRAM.
Expertise firm Meta launched Muse Glimmer, a 30-billion-parameter agentic AI mannequin distributed beneath the permissive Apache 2.0 license. Developed by Meta Superintelligence Labs, the mannequin is designed to function as a totally succesful autonomous agent—together with planning, device invocation, self-verification, and failure restoration—whereas remaining compact sufficient to run regionally on shopper {hardware} with as little as 24 GB of video reminiscence.Â
The weights can be found instantly on Hugging Face, with integrations for widespread inference engines and platforms similar to Ollama, LM Studio, vLLM, SGLang, Collectively AI, Fireworks AI, and OpenRouter scheduled to observe within the coming days.Â
The discharge extends Meta’s custom of open-sourcing foundational AI analysis, this time focusing on the rising demand for native, always-on agent workflows that don’t rely on cloud connectivity or exterior infrastructure.
Muse Glimmer: Structure, Coaching, and Native Optimisation
The brand new AI mannequin was constructed utilizing a bespoke structure and a novel distillation recipe meant to switch agentic reasoning from a considerably bigger instructor mannequin, known as Muse Spark, right into a extra environment friendly type issue. The coaching pipeline comprised three phases: pre-training through logit distillation on the instructor’s outputs; mid-training on extended-context, agent-heavy knowledge enriched with reasoning traces; and post-training combining supervised fine-tuning with on-policy distillation and reinforcement studying throughout normal, coding, and agentic domains. The mannequin was evaluated beneath Meta’s Superior AI Scaling Framework earlier than launch.
Benchmark outcomes point out aggressive efficiency relative to equally sized counterparts, together with Gemma4-31B and Qwen3.6-27B, on duties similar to DeepSearch QA, MCP-Atlas, Ï„-Bench, and SWE-Bench. Past core reasoning, Muse Glimmer helps multimodal enter via a devoted notion encoder, multilingual operation throughout greater than 100 languages, and compatibility with agentic orchestration patterns similar to OpenClaw.
To allow sensible native deployment, Meta utilized quantisation strategies that compress the mannequin to roughly 4-bit precision, decreasing its footprint to beneath 20 GB. This leaves ample reminiscence for the KV cache, picture encoder, and a light-weight speculative decoding drafter based mostly on DFlash, which proposes token blocks in parallel to speed up era with out altering output high quality.Â
Meta validated the setup on MacBook M4-Max, M5-Max, and RTX-5090 {hardware}, reporting speeds appropriate for fluid dialog and real-time agent interplay fully on-device.
Disclaimer
In keeping with the Belief Venture tips, please word that the data offered on this web page is just not meant to be and shouldn’t be interpreted as authorized, tax, funding, monetary, or another type of recommendation. It is very important solely make investments what you’ll be able to afford to lose and to hunt unbiased monetary recommendation you probably have any doubts. For additional info, we propose referring to the phrases and situations in addition to the assistance and assist pages offered by the issuer or advertiser. MetaversePost is dedicated to correct, unbiased reporting, however market situations are topic to alter with out discover.
About The Writer
Alisa, a devoted journalist on the MPost, makes a speciality of crypto, AI, investments, and the expansive realm of Web3. With a eager eye for rising traits and applied sciences, she delivers complete protection to tell and have interaction readers within the ever-evolving panorama of digital finance.
Extra articles
Alisa, a devoted journalist on the MPost, makes a speciality of crypto, AI, investments, and the expansive realm of Web3. With a eager eye for rising traits and applied sciences, she delivers complete protection to tell and have interaction readers within the ever-evolving panorama of digital finance.
Extra articles
