OPSD-Qwen3-8B
Qwen3-8B NoThink post-training checkpoints released as thinking-leakage research artifacts, with a separate community imatrix GGUF conversion.
What happened lately
OPSD-Qwen3-8B timeline
OPSD-Qwen3-8B contains Qwen3-8B checkpoints post-trained in NoThink mode on DAPO-Math for research into thinking leakage. The source model card declares Apache-2.0 and explicitly says the artifacts are not intended as improved general-purpose models. It provides step-25, step-50, step-75 and step-100 branches, with main containing the final checkpoint.
The associated paper was submitted September 23, 2026. The public model repository was created October 22 at 01:27 UTC; mradermacher’s separate imatrix GGUF repository followed at 11:17 UTC. The reviewed conversion revision contains 24 model GGUF files and a separate imatrix file. Availability does not establish that quantized versions preserve the original paper’s activation-level findings. AITrending did not run either checkpoint family.