VibeThinker-3B: Exploring the Frontier of Verifiable Reasoning in Small Language Models

2026-06-14

AI Summary

This work investigates the upper limits of verifiable reasoning capabilities under stringent small-model constraints. Guided by the proposed “parameter compression–coverage hypothesis”—which posits that verifiable reasoning can be compressed into a compact reasoning core, positioning small models as complementary pathways rather than merely efficient substitutes to cutting-edge systems—we develop a 3B-parameter compact model through the Spectrum-to-Signal post-training paradigm. This approach integrates curriculum-based supervised fine-tuning, multi-domain reinforcement learning, and offline self-distillation. The resulting model achieves performance on par with or exceeding that of substantially larger flagship models, attaining scores of 94.3 (scaling to 97.1 at test time) on AIME26, 80.2 Pass@1 on LiveCodeBench v6, 96.1% acceptance rate on recent LeetCode contests, and 93.4 on IFEval.

Abstract

This technical report introduces VibeThinker-3B, a compact dense model with 3B parameters developed to investigate how far verifiable reasoning can be pushed within a strictly small-model regime. Building upon the Spectrum-to-Signal post-training paradigm, we systematically enhance the model through an optimized pipeline that includes curriculum-based supervised fine-tuning, multi-domain reinforcement learning, and offline self-distillation. Experimental evaluations demonstrate that VibeThinker-3B achieves frontier-level performance on highly demanding verifiable tasks. Specifically, it attains a score of 94.3 on AIME26 (improving to 97.1 with claim-level test-time scaling), an 80.2 Pass@1 on LiveCodeBench v6, and exhibits strong out-of-distribution generalization with a 96.1\% acceptance rate on recent unseen LeetCode contests. This effectively places it in the performance band of first-tier reasoning systems, matching or exceeding flagship models that are orders of magnitude larger, such as DeepSeek V3.2, GLM-5, and Gemini 3 Pro. Furthermore, a score of 93.4 on IFEval confirms that this extreme reasoning enhancement does not compromise strict instruction controllability. Extending our previous 1.5B work, these findings motivate the Parametric Compression-Coverage Hypothesis, which views verifiable reasoning as compressible into compact reasoning cores, while open-domain knowledge and general-purpose competence require broad parameter coverage over facts, concepts, and long-tail scenarios. This perspective suggests that compact models are not merely deployment-efficient substitutes, but a complementary path toward frontier-level performance in parameter-dense capability regimes.

Problem

• verifiable reasoning
• small language models
• parameter efficiency
• reasoning performance
• compact models

Innovation

• verifiable reasoning
• small language models
• parametric compression
• self-distillation
• reinforcement learning