Skip to main content
Back to resources
Models

Octen-WebGR-4B

Released on Nov 13, 2026

4.36B-parameter web generative retrieval model based on Qwen3-4B-Base: it generates a two-token semantic identifier constrained to a table of 195.9 million real URLs, so the web index lives in the model's weights. Retrieval needs one GPU with about 10 GB of memory instead of hundreds of GB of vectors. Weights under CC BY-NC 4.0 (non-commercial); SID table and tokenizer under Apache 2.0.

Features

  • Generative retrieval: natural-language query, the real URL as output (two-token SID)
  • Table of 195.9 million real URLs built into the weights
  • ~10 GB GPU memory for retrieval (bf16), ~23 GB host RAM with infer_cpu.py
  • Live demo; technique in the Octen-WebGR report (arXiv 2610.35256)

Tags

Discover Octen-WebGR-4B

Visit the official site to download, learn more and start using Octen-WebGR-4B.

Visit the site

Related resources — Models

Models⚡
DeepSeek V4.1 Flash
Released Sep 10, 2026DeepSeek V4.1 Flash

DeepSeek's MoE model with 552B total parameters and 8-16B active per token. 1M-token context, MIT license: free commercial use. Built for online serving, too large for local use on a single PC.

LLMDeepSeekMoEOpen Source
Models⚡
FLUX.1 [dev]
Released Aug 1, 2024FLUX.2 [klein] 4B/9B

The most advanced open image model: excellent prompt adherence, solid anatomy, full fine-tuning. The standard for serious work.

ImmaginiDiffusionFLUX
Models⚡
FLUX.1 [schnell]
Released Aug 1, 2024FLUX.2 [klein] 4B/9B

Very few steps, high quality, and strong text rendering. The distilled FLUX variant, built for speed.

ImmaginiDiffusionFLUX