Bottleneck Tokens for Unified Multimodal Retrieval

Dimensions overall score 8.0

• Bottleneck Tokens (BToks) and Generative Information Condensation enable decoder-only multimodal LLMs to achieve state-of-the-art unified multimodal retrieval with negligible inference overhead. First, existing methods rely on implicit pooling, which overloads the hidden state of a standard vocabulary token (e.g., ) as the sequence-level representation, a mechanism never designed for information aggregation.problem 35%
  Evidencepartial

  Bottleneck Tokens (BToks) and Generative Information Condensation enable decoder-only multimodal LLMs to achieve state-of-the-art unified multimodal retrieval with negligible inference overhead. First, existing methods rely on implicit pooling, which overloads the hidden state of a standard vocabulary token (e.g., ) as the sequence-level representation, a mechanism never designed for information aggregation.

  Implicationpartial

  Abstract-backed fallback claim; anchored extraction has not materialized a public claim row yet.

  Verificationpartial

  partial

• Adapting decoder-only multimodal large language models (MLLMs) for unified multimodal retrieval faces two structural gaps. First, existing methods rely on implicit pooling, which overloads the hidden state of a standard vocabulary token (e.g., ) as the sequence-level representation, a mechanism never designed for information aggregation.method 35%
  Evidencepartial

  Adapting decoder-only multimodal large language models (MLLMs) for unified multimodal retrieval faces two structural gaps. First, existing methods rely on implicit pooling, which overloads the hidden state of a standard vocabulary token (e.g., ) as the sequence-level representation, a mechanism never designed for information aggregation.

  Implicationpartial

  Abstract-backed fallback claim; anchored extraction has not materialized a public claim row yet.

  Verificationpartial

  partial

• ScienceToStartup currently rates this 8.0/10 on the public viability pass. On MMEB-V2 (78 datasets, 3 modalities, 9 meta-tasks), our approach achieves state-of-the-art among 2B-scale methods under comparable data conditions, attaining an Overall score of 59.0 (+3.6 over VLM2Vec-V2) with substantial gains on semantically demanding tasks (e.g., +12.6 on Video-QA). Code availability is flagged in the production record; the public repository link still needs proof alignment.result 35%
  Evidencepartial

  ScienceToStartup currently rates this 8.0/10 on the public viability pass. On MMEB-V2 (78 datasets, 3 modalities, 9 meta-tasks), our approach achieves state-of-the-art among 2B-scale methods under comparable data conditions, attaining an Overall score of 59.0 (+3.6 over VLM2Vec-V2) with substantial gains on semantically demanding tasks (e.g., +12.6 on Video-QA). Code availability is flagged in the production record; the public repository link still needs proof alignment.

  Implicationpartial

  Abstract-backed fallback claim; anchored extraction has not materialized a public claim row yet.

  Verificationpartial

  partial

• Multimodal Retrieval moved forward this cycle; last verified April 2026. Public score 8.0/10. Production flags indicate code availability.why now 35%
  Evidencepartial

  Multimodal Retrieval moved forward this cycle; last verified April 2026. Public score 8.0/10. Production flags indicate code availability.

  Implicationpartial

  Abstract-backed fallback claim; anchored extraction has not materialized a public claim row yet.

  Verificationpartial

  partial

Competitive landscape

Bottleneck Tokens (BToks) and Generative Information Condensation enable decoder-only multimodal LLMs to achieve state-of-the-art unified multimodal retrieval with negligible inference overhead.