← Back to all news
arXiv Computation and Language

Concise-9B: Removing the Slack from a Reasoning Model's Thinking

The paper introduces Concise-9B, a version of the Qwen3.5-9B reasoning model that reduces unnecessary thinking tokens through concise self‑training. By applying two rounds of on‑policy rejection‑sampling fine‑tuning and a small share of answer‑synthesis targets, Concise-9B cuts token usage to roughly two‑thirds of the original model while maintaining comparable accuracy on a wide range of hard mathematics and reasoning benchmarks. The study also shows that most of the token savings occur after the answer has already formed, and that the approach’s effectiveness depends on the verbosity of the base model.

Related stories

arXiv AI
1d ago

Rules, Traces and Routing Statistics: Training and Calibration Data for Open Models on ARC-AGI-2

The paper reports on training and calibration experiments for three open models on the ARC‑AGI‑2 benchmark, using offline runs on four L4 GPUs for up to 12 hours. For the NVARC Qwen3‑4B grid model, a curated corpus of 12,000 executable rules and 95,993 tasks was built, and fine‑tuning on 1,920 real task records raised the semi‑private score from 28.96 to 33.06, with no synthetic‑task checkpoint outperforming the real‑task model. For the Qwen3.8‑27B reasoning model, fine‑tuning on 1,036 of its own reasoning traces and a specialized inference procedure increased the score from 26.81 to 44.03, while the Qwen3.8‑Flash‑Next mixture‑of‑experts model achieved a best score of 56.67 after expert pruning, routing statistics adaptation, and an FP8 indexer cache.

By Nguyen Minh Thang, Trinh Thi Phuong
fine-tuningefficiency
More like this →
arXiv AI
6d ago

ThinkJev: One Thought, Every Budget for Calibrated Typed Decisions

ThinkJev introduces a hybrid approach that lets a language model decide whether to answer directly or to think first, using the same underlying model for both fast and slow responses. By reading the model’s probability distribution over option letters after any number of thought tokens, the system can provide a calibrated decision under any thinking budget without retraining. Experiments with Qwen3.5-27B show that the method achieves high accuracy while reducing the number of thought tokens needed, and it performs competitively on the JevBench benchmark.

By Xiaolu Ding, Kai Yang, Dongdong Liu
llmsbenchmarks
More like this →
arXiv AI
Aug 21

Learning When to Think: Adaptive Reasoning for Test-Time Compute Allocation

arXiv:2608. 20256v1 Announce Type: new Abstract: Reasoning language models trained with reinforcement learning typically operate under a fixed token budget rather than an explicitly adaptive one, which can lead to over-computation on easy problems and insufficient computation on difficult ones.

By Gijs Kassenaar, Zhao Yang, Vincent Fran\c{c}ois-Lavet
llmsreinforcement-learningbenchmarks
More like this →
arXiv Machine Learning
Oct 15

Inverse Reasoning Models for Efficient Post-Training and Verification from Answer-Only Data

The paper introduces a self‑distillation method that leverages answer‑only data to improve large language models (LLMs) after training. By training an inverse reasoning LLM to generate latent chain‑of‑thought traces conditioned on the answer, the authors can perform offline distillation that significantly boosts performance on hard prompts for tasks like Countdown and GSM8K. Additionally, the inverse model serves as an accurate verifier, enhancing best‑of‑n selection accuracy.

By Eric O. Korman
llmsreinforcement-learningfine-tuningefficiency
More like this →