Daniel Han's Unsloth says the new quants beat GSQ-RCO by 10-15% on agentic coding tests, with swappable engrams from 28.8 GB to 102.4 GB.
By [Ryan Merket](https://runtimewire.com/author/ryan-merket)
· Published
· Updated
Primary source: [x.com](https://x.com/danielhanchen/status/2118037508139413742)
Why it matters #
Quantization affects whether a 125B-parameter model can run locally. Unsloth's engram switching gives users a direct trade between file size and measured error.
Unsloth co-founder Daniel Han (@danielhanchen) released new Dynamic v3.1 GGUFs for Qwen3.8-Flash-Next on November 4th, saying the files score 10-15% higher on DeepSWE, Terminal Bench, Aider Polyglot and other tests than GSQ-RCO in Unsloth's own benchmarking.…
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.
source & further reading
runtimewire.com — original article
Enrich Labs launches a tracker for whether shopping agents pick your brand
Chen-Ping Yu launches Atmee v2, a real-time avatar model from one image
Confluent makes Lightning Tables generally available for querying streaming data