LessThink-Qwen3-4B
A reasoning model built on Qwen3-4B through post-training optimization. It reduces reasoning token consumption by approximately 44% by compressing chain-of-thought length, while preserving the base model's knowledge base and response style.
Timeline (last 90 days)
LessThink-Qwen3-4B is a model obtained by post-training Qwen3-4B to reduce token consumption by approximately 44% during the inference stage while preserving the original knowledge coverage and response style as much as possible.
The entire training pipeline of LessThink-Qwen3-4B was completed using only a single GPU.
LessThink-Qwen3-4B was released by a developer on Reddit without changing the base model or increasing the number of parameters.
The 44% reduction in inference tokens is a self-reported figure from the author and has not been verified by third-party reproduction; the original materials do not disclose the data construction method, training algorithm, sampling scale, or benchmark test details.
Training thinking-length preferences into the model weights is more stable than using system prompts at deployment time to request concise answers, as the latter is an external constraint that is unstable while the former becomes the model's default behavior.
All Facts (5)
LessThink-Qwen3-4B is a model obtained by post-training Qwen3-4B to reduce token consumption by approximately 44% during the inference stage while preserving the original knowledge coverage and response style as much as possible.
50%UnverifiedThe entire training pipeline of LessThink-Qwen3-4B was completed using only a single GPU.
50%UnverifiedThe 44% reduction in inference tokens is a self-reported figure from the author and has not been verified by third-party reproduction; the original materials do not disclose the data construction method, training algorithm, sampling scale, or benchmark test details.
50%UnverifiedTraining thinking-length preferences into the model weights is more stable than using system prompts at deployment time to request concise answers, as the latter is an external constraint that is unstable while the former becomes the model's default behavior.
50%UnverifiedLessThink-Qwen3-4B was released by a developer on Reddit without changing the base model or increasing the number of parameters.
50%