Paper 2510.27492
ThinkMorph: Emergent Properties in Multimodal Interleaved Chain-of-Thought Reasoning
- Published
- Oct 2025
- Research lab
- Independent
- Citations
- 40
- GitHub
- 192 stars
01 In brief
Summary
ThinkMorph is a unified multimodal model fine-tuned on ~24K interleaved reasoning traces across four tasks (Jigsaw Assembly, Spatial Navigation, Visual Search, Chart Refocus) to enable complementary text-image chain-of-thought reasoning.
It outperforms its base model Bagel-7B by an average of 34.74% on vision-centric benchmarks, with gains of 85.84% on Spatial Navigation and 38.75% on Jigsaw Assembly.
Interleaved reasoning surpasses text-only and visual-only modes by 5.33% on average.
ThinkMorph generalizes out-of-domain, beating InternVL3.5-38B on SAT (52.67% vs.
49.33%) and matching Gemini 2.5 Flash on MMVP (80.33%).
Three emergent properties are identified: (1) unseen visual manipulations (e.g., zoom-in, inpainting) that aid reasoning; (2) autonomous mode switching, where the model adaptively uses text-only reasoning in 5.3% of cases, improving accuracy by 7.29% and reducing token use by ~75%; (3) better test-time scaling via diversified thoughts, yielding gains like +8.0% on BLINK-J under Best-of-N sampling.
The model also shows that interleaved reasoning enriches text representations with higher entropy and vocabulary diversity.
02 From the paper
Abstract
Multimodal reasoning requires iterative coordination between language and vision, yet it remains unclear what constitutes a meaningful interleaved chain of thought. We posit that text and image thoughts should function as complementary rather than isomorphic modalities that mutually advance reasoning. Guided by this principle, we build ThinkMorph, a unified model fine-tuned on approximately 24K high-quality interleaved reasoning traces spanning tasks with varying visual engagement. ThinkMorph learns to generate progressive text-image reasoning steps that concretely manipulate visual content while maintaining coherent verbal logic. It delivers large gains on vision-centric benchmarks (averaging 34.7 percent over the base model) and generalizes to out-of-domain tasks, matching or surpassing larger and proprietary VLMs. Beyond performance, ThinkMorph exhibits emergent multimodal intelligence, including unseen visual manipulation skills, adaptive switching between reasoning modes, and better test-time scaling through diversified multimodal thoughts. These findings suggest promising directions for characterizing the emergent capabilities of unified models for multimodal reasoning.