# LBS-27B

> LBS-27B is an agentic vision-language model based on Qwen3.8-27b. It learns to select and inspect image regions through GUI grounding, then transfers this observation policy to general-domain visual question answering and high-resolution perception.

- **Organisation:** inclusionAI
- **Category:** Vision-Language Models
- **Parameters:** 27B
- **Context length:** 262,144 tokens
- **Licence:** Apache 2.0 (Commercial use OK)
- **Released:** 2026-10-26
- **Hugging Face:** 2K downloads (30 days), 16 likes

## Estimated memory

| Quant | ≈ GB |
|---|---|
| Q2_K | 11.18 |
| Q3_K_M | 15.4 |
| Q4_K_M | 18.51 |
| Q5_K_M | 21.28 |
| Q6_K | 24.25 |
| Q8_0 | 30.55 |
| F16 | 55.31 |

## Runs with

- [llama.cpp](https://selfhost.directory/project/llamacpp)
- [SGLang](https://selfhost.directory/project/sglang)
- [vLLM](https://selfhost.directory/project/vllm)

## About

LBS-27B is an agentic vision-language model based on Qwen3.8-27b. It learns to select and inspect image regions through GUI grounding, then transfers this observation policy to general-domain visual question answering and high-resolution perception.

Training uses only synthetic GUI grounding data, with no general-domain data or VQA questions: approximately 10K cold-start SFT examples, followed by 80 GRPO steps with a batch size of 64 and 8 rollouts per prompt.

• Adaptive observation: zoom into relevant regions, refine previous crops,
• and revisit the original image when needed.
• Compositional zoom-in: request up to two independent crops per turn,
• with a total budget of three crop calls per episode.
• Transfer beyond screens: apply the learned observation policy to visual
• QA without general-domain VQA training.

All LBS scores use agentic zoom-in over complete benchmark runs. PerceptionBench uses a self-hosted Deepseekv4-flash-0731 judge, with zero failed or skipped judgments; repeated judgments may vary. Chart references use different inference setups and GPT-oss-120B for PerceptionBench, so comparisons are not controlled. See comparison notes.

More result visualizations are available in the LBS project; we have not performed any additional fine-tuning on general-domain data and welcome further RL experiments in specific subdomains.

## Alternatives

- [Qwen/Qwen3.8-27B](https://selfhost.directory/models/qwen/qwen3.8-27b) (27.8B, Apache 2.0, ≈18.97 GB at Q4): Following the widespread community adoption of the Qwen3.5 and Qwen3.6 series, we are pleased to introduce Qwen3.8, the most capable generation in the Qwen open-model family to date.
- [google/gemma-4-26B-A4B-it](https://selfhost.directory/models/google/gemma-4-26b-a4b-it) (25.8B (4B active), Apache 2.0, ≈17.67 GB at Q4): Gemma is a family of open models built by Google DeepMind. Gemma 4 models are multimodal, handling text and image input (with audio supported on E2B, E4B, and 12B) and generating text output. This release includes open-weights models in both pre-trained and instruction-tuned variants.
- [Qwen/Qwen3.6-27B](https://selfhost.directory/models/qwen/qwen3.6-27b) (27.8B, Apache 2.0, ≈18.97 GB at Q4): Following the February release of the Qwen3.5 series, we're pleased to share the first open-weight variant of Qwen3.6. Built on direct feedback from the community, Qwen3.6 prioritizes stability and real-world utility, offering developers a more intuitive, responsive, and genuinely productive coding…
- [google/gemma-4-31B-it](https://selfhost.directory/models/google/gemma-4-31b-it) (31.3B, Apache 2.0, ≈26.54 GB at Q4): Gemma is a family of open models built by Google DeepMind. Gemma 4 models are multimodal, handling text and image input (with audio supported on E2B, E4B, and 12B) and generating text output. This release includes open-weights models in both pre-trained and instruction-tuned variants.
- [Qwen/Qwen3.5-27B](https://selfhost.directory/models/qwen/qwen3.5-27b) (27.8B, Apache 2.0, ≈18.97 GB at Q4): Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance.
- [google/diffusiongemma-26B-A4B-it](https://selfhost.directory/models/google/diffusiongemma-26b-a4b-it) (25.8B (4B active), Apache 2.0, ≈17.68 GB at Q4): DiffusionGemma is a generative model built by Google DeepMind. Based on the 26B A4B Mixture-of-Experts (MoE) Gemma 4 architecture, DiffusionGemma generates tokens using discrete diffusion. This open-weights model is multimodal, handling text, image, and video inputs to generate text output.

More: https://selfhost.directory/models/inclusionai/lbs-27b/alternatives

## Runs well on

- [Tesla P40](https://selfhost.directory/models/inclusionai/lbs-27b/on/tesla-p40): Q5_K_M, est. 9.9 tok/s
- [Radeon RX 7900 XTX](https://selfhost.directory/models/inclusionai/lbs-27b/on/rx-7900-xtx): Q5_K_M, est. 27 tok/s
- [Instinct MI50 32GB](https://selfhost.directory/models/inclusionai/lbs-27b/on/instinct-mi50): Q6_K, est. 25 tok/s
- [GeForce RTX 5090](https://selfhost.directory/models/inclusionai/lbs-27b/on/rtx-5090): Q6_K, est. 44 tok/s
- [Apple M4 Pro (48 GB)](https://selfhost.directory/models/inclusionai/lbs-27b/on/apple-m4-pro-48gb): Q8_0, est. 5.2 tok/s
- [RTX A6000](https://selfhost.directory/models/inclusionai/lbs-27b/on/rtx-a6000): Q8_0, est. 15 tok/s

## Similar size

- [Qwen/Qwen3.8-27B](https://selfhost.directory/models/qwen/qwen3.8-27b) (27.8B)
- [google/gemma-4-31B-it](https://selfhost.directory/models/google/gemma-4-31b-it) (31.3B)
- [google/gemma-4-26B-A4B-it](https://selfhost.directory/models/google/gemma-4-26b-a4b-it) (25.8B)
- [Qwen/Qwen3.6-27B](https://selfhost.directory/models/qwen/qwen3.6-27b) (27.8B)
- [google/gemma-3-27b-it](https://selfhost.directory/models/google/gemma-3-27b-it) (27.4B)
- [meta-models/Muse-Glimmer-30B](https://selfhost.directory/models/meta-models/muse-glimmer-30b) (29.8B)

## FAQ

**How much VRAM does LBS-27B need?**
About 19 GB at Q4_K_M with an 8K context. Q8_0 needs about 31 GB and full precision about 55 GB. Longer contexts need more memory for the KV cache.

**Can LBS-27B run on the Tesla P40?**
Yes. It fits at Q5_K_M (about 23 GB), at an estimated 9.9 tokens/s. It fits comfortably on the Radeon RX 7900 XTX, Instinct MI50 32GB, RTX 5090.

**Can I use LBS-27B commercially?**
Yes. It is released under Apache 2.0. Permissive: use, modify and sell freely; keep the notice and licence text.

**What is the context length of LBS-27B?**
262,144 tokens (256K). The full window needs far more memory than an 8K context.

**How do I run LBS-27B with llama.cpp?**
Run llama-server -hf bartowski/LBS-27B-GGUF:Q4_K_M. It downloads the GGUF and starts an OpenAI-compatible server.

## Links

- huggingface: https://huggingface.co/inclusionAI/LBS-27B
- github: https://github.com/inclusionAI/UI-Venus

---
Source: https://selfhost.directory/models/inclusionai/lbs-27b