The Synergy Dilemma of Long-CoT SFT and RL: Investigating Post-Training Techniques for Reasoning VLMs

Large vision-language models (VLMs) increasingly adopt post-training techniques such as long chain-of-thought (CoT) supervised fine-tuning (SFT) and reinforcement learning (RL) to elicit sophisticated reasoning. While these methods exhibit synergy in language-only models, their joint effectiveness in VLMs remains uncertain. We present a systematic investigation into the distinct roles and interplay of long-CoT SFT and RL across multiple multimodal reasoning benchmarks. We find that SFT improves performance on difficult questions by in-depth, structured reasoning, but introduces verbosity and degrades performance on simpler ones. In contrast, RL promotes generalization and brevity, yielding consistent improvements across all difficulty levels, though the improvements on the hardest questions are less prominent compared to SFT. Surprisingly, combining them through two-staged, interleaved, or progressive training strategies, as well as data mixing and model merging, all fails to produce additive benefits, instead leading to trade-offs in accuracy, reasoning style, and response length. This "synergy dilemma" highlights the need for more seamless and adaptive approaches to unlock the full potential of combined post-training techniques for reasoning VLMs.Figure 1: Accuracy gains from various post-training techniques across five difficulty levels (L1, easy to L5, hard) on five multimodal reasoning benchmarks. Long-CoT SFT boosts Qwen2.5-VL-7B on harder questions but hurts easier ones, while RL yields steady gains across the board. Hybrid strategies consistently trade off strengths rather than achieving true synergy. Large language models (LLMs) like OpenAI's o1/o3 (Jaech et al., 2024) and DeepSeek-R1 (Guo et al., 2025) have demonstrated remarkable reasoning abilities by thinking before answering .