# Router-Suggest: Dynamic Routing for Multimodal Auto-Completion in Visually-Grounded Dialogs

Source: [Router-Suggest: Dynamic Routing for Multimodal Auto-Completion in Visually-Grounded Dialogs](<https://www.sota2.com/research/paper/69647fda53fdd5c3617445a9>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-01-15T06:00:02.208Z

Authors: Sandeep Mishra, Devichand Budagam, Anubhab Mandal, Bishal Santra, Pawan Goyal, Manish Gupta

Published: 2026-01-09T15:29:50.000Z

## Abstract

Real-time multimodal auto-completion is essential for digital assistants, chatbots, design tools, and healthcare consultations, where user inputs rely on shared visual context. We introduce Multimodal Auto-Completion (MAC), a task that predicts upcoming characters in live chats using partially typed text and visual cues. Unlike traditional text-only auto-completion (TAC), MAC grounds predictions in multimodal context to better capture user intent. To enable this task, we adapt MMDialog and ImageChat to create benchmark datasets. We evaluate leading vision-language models (VLMs) against strong textual baselines, highlighting trade-offs in accuracy and efficiency. We present Router-Suggest, a router framework that dynamically selects between textual models and VLMs based on dialog context, along with a lightweight variant for resource-constrained environments. Router-Suggest achieves a 2.3x to 10x speedup over the best-performing VLM. A user study shows that VLMs significantly excel over textual models on user satisfaction, notably saving user typing effort and improving the quality of completions in multi-turn conversations. These findings underscore the need for multimodal context in auto-completions, leading to smarter, user-aware assistants.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Multimodal Autocomplete on MMDD unseen prefixes](<https://www.sota2.com/research/sota/multimodal-autocomplete-on-mmdd-unseen-prefixes.md>): Multimodal Autocomplete; MMDD unseen prefixes; PR-F1: 35.6.

- [Multimodal Autocomplete on ImageChat (unseen prefixes)](<https://www.sota2.com/research/sota/multimodal-autocomplete-on-imagechat-unseen-prefixes.md>): Multimodal Autocomplete; ImageChat (unseen prefixes); PR-F1: 0.281.

- [Multimodal Dialogue Response Generation on MMDD seen prefixes (test)](<https://www.sota2.com/research/sota/multimodal-dialogue-response-generation-on-mmdd-seen-prefixes-test.md>): Multimodal Dialogue Response Generation; MMDD seen prefixes (test); TR: not recorded.

- [Multimodal Dialogue Response Generation on ImageChat seen prefixes (test)](<https://www.sota2.com/research/sota/multimodal-dialogue-response-generation-on-imagechat-seen-prefixes-test.md>): Multimodal Dialogue Response Generation; ImageChat seen prefixes (test); TR: not recorded.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2601.05851>)

[PDF](<https://arxiv.org/pdf/2601.05851.pdf>)

[Implementation](<https://github.com/devichand579/MAC>)
