RecGPT-V3 Technical Report

Hugging Face Daily Papers Papers
recommender-system large-language-models stateful hybrid-modal memory-hub latent-reasoning efficiency

Summary

RecGPT-V3 introduces a stateful, hybrid-modal recommender that uses a Memory Hub and Latent Intent Reasoning to cut computation by 55.8% and output token cost by 200x, achieving significant gains in Taobao's feed.

Large language models (LLMs) are transforming recommender systems from matching co-occurrence patterns in historical behavior toward reasoning about the intent that drives it. RecGPT-V1 pioneered this paradigm on Taobao by centering user understanding, and RecGPT-V2 scaled it via coordinated multi-agent reasoning; both are deployed in production with consistent gains in user experience and commercial outcomes. However, operating RecGPT at scale reveals three challenges: (1) stateless behavior modeling, where each request reprocesses full user history, wasting computation and discarding prior analysis; (2) a tag-to-item information bottleneck, where natural-language tags form a lossy channel between user understanding and item grounding; and (3) inefficient explicit reasoning, whose lengthy chain-of-thought incurs untenable latency and compute overhead. We present RecGPT-V3, a stateful, hybrid-modal recommender that reasons over natural language for open-world knowledge and Semantic IDs (SIDs) for concrete item grounding. A Memory Hub maintains structured, continually evolving user memory that distills long-horizon behavior into condensed units, cutting user-modeling computation by 55.8%. A Hybrid-modal Foundation Model allows the LLM jointly reason over text tags and SIDs, opening a high-bandwidth channel into the item space. Latent Intent Reasoning internalizes verbose rationales into compact learnable latent tokens that remain decodable into readable explanations, lowering output token cost by 200x. Deployed in Taobao's "Guess What You Like" feed, RecGPT-V3 achieves consistent gains in large-scale online A/B tests: IPV +1.28%, CTR +1.00%, TC +1.97%, GMV +3.97%, while cutting end-to-end serving resource consumption by 52.4%.
Original Article
View Cached Full Text

Cached at: 07/20/26, 09:41 AM

Paper page - RecGPT-V3 Technical Report

Source: https://huggingface.co/papers/2607.15591 Authors:

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

,

Abstract

Largelanguagemodels(LLMs)aretransformingrecommendersystemsfrommatchingco-occurrencepatternsinhistoricalbehaviortowardreasoningabouttheintentthatdrivesit.RecGPT-V1pioneeredthisparadigmonTaobaobycenteringuserunderstanding,andRecGPT-V2scaleditviacoordinatedmulti-agentreasoning;botharedeployedinproductionwithconsistentgainsinuserexperienceandcommercialoutcomes.However,operatingRecGPTatscalerevealsthreechallenges:(1)statelessbehaviormodeling,whereeachrequestreprocessesfulluserhistory,wastingcomputationanddiscardingprioranalysis;(2)atag-to-iteminformationbottleneck,wherenatural-languagetagsformalossychannelbetweenuserunderstandinganditemgrounding;and(3)inefficientexplicitreasoning,whoselengthychain-of-thoughtincursuntenablelatencyandcomputeoverhead.WepresentRecGPT-V3,astateful,hybrid-modalrecommenderthatreasonsovernaturallanguageforopen-worldknowledgeandSemanticIDs(SIDs)forconcreteitemgrounding.AMemoryHubmaintainsstructured,continuallyevolvingusermemorythatdistillslong-horizonbehaviorintocondensedunits,cuttinguser-modelingcomputationby55.8%.AHybrid-modalFoundationModelallowstheLLMjointlyreasonovertexttagsandSIDs,openingahigh-bandwidthchannelintotheitemspace.LatentIntentReasoninginternalizesverboserationalesintocompactlearnablelatenttokensthatremaindecodableintoreadableexplanations,loweringoutputtokencostby200x.DeployedinTaobao’s“GuessWhatYouLike“feed,RecGPT-V3achievesconsistentgainsinlarge-scaleonlineA/Btests:IPV+1.28%,CTR+1.00%,TC+1.97%,GMV+3.97%,whilecuttingend-to-endservingresourceconsumptionby52.4%.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2607\.15591

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2607.15591 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2607.15591 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2607.15591 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

GPT-3.5 Turbo fine-tuning and API updates

OpenAI Blog

OpenAI has released fine-tuning capabilities for GPT-3.5 Turbo, allowing developers to customize models for specific use cases with improved performance, steerability, and output formatting. The update enables fine-tuned GPT-3.5 Turbo to match GPT-4 performance on certain tasks while reducing prompt sizes by up to 90%.

Customizing GPT-3 for your application

OpenAI Blog

OpenAI has launched fine-tuning capabilities for GPT-3, allowing developers to customize the model on their own data via a single CLI command, resulting in improved accuracy, reduced costs, and lower latency for production use cases. Early customers like Keeper Tax, Viable, and Sana Labs report significant accuracy improvements after fine-tuning.