Free(): Learning to Forget in Malloc-Only Reasoning Models

TL;DR

Free()LM introduces a self-forgetting mechanism in reasoning models, enabling dynamic pruning of obsolete information to improve performance and stability across various model scales and tasks.

Contribution

It proposes the Free-Module, a plug-and-play adapter that allows models to identify and prune useless context, addressing the flaw of information accumulation in standard LLMs.

Findings

Achieves 3.3% average improvement over top reasoning baselines.

Establishes new SOTA on IMOanswerBench with DeepSeek V3.2-Speciale.

Restores performance from 0% to 50% accuracy in long-horizon tasks.

Abstract

Reasoning models enhance problem-solving by scaling test-time compute, yet they face a critical paradox: excessive thinking tokens often degrade performance rather than improve it. We attribute this to a fundamental architectural flaw: standard LLMs operate as "malloc-only" engines, continuously accumulating valid and redundant steps alike without a mechanism to prune obsolete information. To break this cycle, we propose Free()LM, a model that introduces an intrinsic self-forgetting capability via the Free-Module, a plug-and-play LoRA adapter. By iteratively switching between reasoning and cleaning modes, Free()LM dynamically identifies and prunes useless context chunks, maintaining a compact and noise-free state. Extensive experiments show that Free()LM provides consistent improvements across all model scales (8B to 685B). It achieves a 3.3% average improvement over top-tier reasoning baselines, even establishing a new SOTA on IMOanswerBench using DeepSeek V3.2-Speciale. Most notably, in long-horizon tasks where the standard Qwen3-235B-A22B model suffers a total collapse (0% accuracy), Free()LM restores performance to 50%. Our findings suggest that sustainable intelligence requires the freedom to forget as much as the power to think.