It’s out!
What does this active parameters business about? Is it supposed to perform similar to much bigger models at the same RAM usage?
It’s an MoE (Mixture of Experts) approach. An 80B-A3B model has 80B parameters total, so that dictates the size of the model and the VRAM+RAM you need to have to hold it, but only 3B of those parameters are active at any given time. This reduces the intelligence of the model compared to an 80B dense model, but improves the speed. In the end it’s the size of an 80B model, with the intelligence of a ~40B model, that runs at the speed of a 3B model.
Pretty much all state of the art models either have already, or are in the process of switching to an MoE design, since it significantly reduces the hardware required to run big models at usable speeds. You can often get usable speeds on MoEs without a GPU at all.
Interesting. 3B models run decently fast on my CPU and I have a lot of system RAM. 🤔
E: Just tried it on 100% CPU on AMD 7700 with DDR5 3600 and it does 6.5t/s. Not bad.
As far as I know fewer active parameters means faster. There’s less arithmetic calculations to be done per pass. But all parameters need to be kept in memory, because they might become active the next pass. So it won’t save any RAM.
They have a short paragraph in the description. It has 80B total parameters, 3B active each pass. It achieves performance like a 30-60B model (10-20x, their claim). But is way more efficiant than that with only 3B active parameters.
Got. Thanks!
Looks like a solid model based on my limited testing. Though tool calls frequently fail with “JSON parse error” in longer sessions in OpenCode and llama.cpp. Hoping that will be addressed soon.
Yeah I enjoy it as well. Just in case you missed it - a fix was merged into llama.cpp two days ago which is said to improve quality.
Edit: I stand corrected - the fix for the issue you’re experiencing has not yet been merged.
thanks to the beauty of mixture-of-experte models i can shove a q2 quant of this into my 8gb vram
Seems pretty good, using the latest version of ollama (downloaded the default Q4 from ollama) and then popped it into Codex with this config.toml:
model = "qwen3-coder-next:Q4_K_M" model_provider = "ollama" model_reasoning_effort = "medium" [model_providers.ollama] name = "Ollama" base_url = "http://localhost:11434/v1" [analytics] enabled = falseWorks well in Codex CLI and VScode Codex IDE plugin. Did not work well with Kilo Code or Roo plugins unfortunately (but I have yet to find much that does).
I am not an expert, this may not be the best way, I don’t know… just sharing my experience for the other non-experts out there.
I had trouble with two Unsloth quants and had to switch to Bartowski’s quant.
IMO it’s a very good model, not just for coding. It’s also very good as a general model. I might even prefer it to instruct.
