Meta Muse Glimmer 30B

“Muse Glimmer is better in practice on low reasoning strength, and while its token generation is slow, it actually gets to answers pretty quickly.”

“Muse Glimmer (dense 30B), in practice, can get to better answers more quickly than Qwen 3.6 35B-A3B.”

“I really like Muse Glimmer, and on low reasoning level it's both helpful and quite quick.”

“Muse Glimmer sometimes decides it has what it needs and has no need to ask”

“runs quite well on my M1 Max as it is 30B but it also reasons very efficiently”

“Muse Glimmer may be a little slower per token, but it uses them very efficiently”

“you will likely find it solves actual problems faster than Qwen 3.6 35B-A3B”

“Muse Glimmer is fine, actually. Gets to the solution as quickly as Qwen 3.6 35B-A3B.”

“I found it more usable on my hardware because I can get an answer quicker.”

“holds its own (for agentic code workloads) against models 5x to 10x its size”

“I like the style of glimmer more. Much terser language, no adjectives, no fluffy claude-like language.”

“It gets the job done quicker than A3B despite being several times slower.”

“It is quite good, far better than the benchmarks let on, especially in heavily agentic cases”

“Muse Glimmer actually has by far the lowest per-token cache usage for the competitive 30B-class dense models”

“I'm in the exact same boat with a 7900 XT and a good Glimmer 30B experience”

“I was quite impressed by Muse Glimmer, and while I am sure people will observe that it is less good on benchmarks”

“I was getting some decent results out of Muse Glimmer which was more than twice as fast”

“I can run glimmer with a full 128k context window with the draft model at 65-80 tok/s.”

“Like a talented, slightly snarky guy who is maybe a bit of a dick but quite fun to be around.”

“we've optimised our inference engine to run Meta Muse Glimmer 30B faster than any other engine”

“They were all able to build a working collaborative whiteboard app, without any guidance”

“I have it running sometimes at least 8 hours straight on a single prompt and no weirdness.”

“Optimizing models specifically for always-on local agent workflows is the right primitive”

“for FOSS and self hosted it is good - but I am still wondering how are they going to Meta-stasize it”

“wasn't impressed, gave it some defective code, and asked it to fix all errors. It kept looping around and around”

“Having spent a good part of the day with it, glimmer reminds me of Rorschach from The Watchmen. No unessential parts of speech, action oriented, brief and to the point.”

“I agree with their open-source model approach, but actually trusting Meta… to protect my privacy and my data”

“I haven't really got that far in, but it writes in a sort of clipped, geeky note form in the reasoning traces”

“its thinking traces are really quite terse. It genuinely seems to spend many fewer tokens on reasoning”

“But then Glimmer already loses to 3.6 27B on it by a large margin.”

“It wasn't super fast, but it was manageable and it completed the tasks I gave it well.”

“It has all of the components of a pelican riding a bicycle, though not exactly arranged in the right order!”

“I kind of prefer this workflow, as I stay more in touch with how the codebase is structured.”

“seems to underperform on Terminal Bench compared with qwen3.6-27b: 51.7 vs 60.7”

“Given Meta's track record, I would not touch anything from them with a 10 feet pole.”

“Glad to see the extra engineering effort that went into creating this local model and making it run well on a consumer device.”

“This release is not a meaningful improvement in any metric over 5 months old Qwen 3.6.”

“Glimmer looks really strong, I'm encouraged that Meta compared it to 3.6 in the model card!”

“I'm glad it's open. It does not make me think any better of Meta.”

“This model I think will be too slow for that on Spark, even at 4 bit quant.”