LBS-27B

# Fix list: LBS-27B

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/lbs-27b, the October 2026 research run, assessed 27 October 2026. Grade D, 53.9 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on LBS-27B: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Reach, 55 out of 100, up to 17.3 more on the total

Why it scored 55: Memory earns 10, since the only build is 54.73 GB of BF16, about 66 GB with a fifth added, which fits the reference machine but not 64 GB. Runtimes earn 20 from released vLLM, SGLang, Transformers and llama.cpp, which all carry Qwen3.8-27B's architecture, and Ollama and LM Studio have no build. Builds earn 0, as inclusionAI publishes no quantised build. Terms earn 15 for Apache 2.0 with nothing added, on the weights and on the base. Download earns 10, with no gate. Hosted earns 0, as no provider or inclusionAI API serves it.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reach):

Whether an agent can run it, and where.

- 0 to 30, memory. The smallest build from the publisher or a runtime's own library that keeps 4-bit precision or better, at its file size plus a fifth for the runtime and a working context. 30 for 16 GB or less, 25 for 24 GB, 20 for 32 GB, 15 for 64 GB, 10 for what fits our reference machine (the GPU share of 128 GB of unified memory), 5 for 256 GB, 0 above.
- 0 to 25, runtimes. 5 for each of llama.cpp, Ollama, vLLM, SGLang, LM Studio and Hugging Face Transformers that runs the architecture in a released version, from the runtime's own code, library or documentation. A publisher's fork or an unreleased branch doesn't count. 25 at most.
- 0 to 15, builds from the publisher. 15 for quantised builds at two precisions or more (FP8, 4-bit, GGUF), 10 for one, 0 when only other people's quantisations exist.
- 0 to 15, terms of use. 15 for an OSI-approved licence with nothing added, 10 for a custom licence that lets anyone use it commercially for free below a threshold or under a use policy, 5 for research or non-commercial use only, 0 for no licence or unclear terms.
- 0 to 10, download. 10 when the files download with no account and no terms to accept, 5 for a click-through that needs an account, 0 when a person approves each request.
- 5, hosted too. The publisher's own API serves it, or at least two inference providers do, so an agent without the hardware can still call it.

When our reference-machine runs start, speed on that machine (tokens per second and time to first token at the build that fits) joins Reach.

## 2. Risk, 50 out of 100, up to 15.4 more on the total

Why it scored 50: File safety earns 15, safetensors only but with the documented route running the publisher's scripts to serve it and to write its system prompt. Provenance earns 15, inclusionAI's own organisation with a hash for every file. Safety evaluation and agent guidance earn 0, as nothing is published on either, prompt injection through screenshots included. Licence risk keeps all 15 under Apache 2.0. The security programme earns 5 for Ant Group's AntSRC, which isn't linked to the model.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-risk):

What could go wrong when you run it, and what the publisher has done about it. A higher score means less risk.

- 0 to 25, file safety. 25 for weights in safetensors or GGUF only, 15 when pickle files (.bin, .pt, .pth) sit beside them, 0 for pickle only. 10 off when the documented way to load it or to build its prompts runs code from the repository (`trust_remote_code`, or the publisher's scripts in place of a template).
- 0 to 15, provenance. 10 when the files come from the publisher's own organisation, the one its site or docs link to, and 5 for a checksum on every file.
- 0 to 20, safety evaluation. 20 when results are published and someone other than the publisher ran them or can rerun them, 10 for the publisher's own results, 0 for none.
- 0 to 15, guidance for agent use. 15 for documented prompt-injection guidance or an instruction hierarchy the model was trained to follow, 10 for a statement of what its trained refusals cover and the safeguards to add around it, 5 for general cautions, 0 for none.
- 0 to 15, licence risk. 15 to start, 5 off for each of these, a use policy or field-of-use limits beyond the law, terms the licensor can change or revoke, a revenue or user threshold, and a licence that changed without a dated note or whose link is broken. Not below 0.
- 0 to 10, a security programme. 5 for a way to report a flaw (security.txt, a disclosure policy or a security contact), 5 for advisories handled in public.

When our reference-machine runs start, measured robustness (prompt injection with AgentDojo, over-refusal with XSTest) joins Risk.

## 3. Documentation and upkeep, 43 out of 100, up to 8.8 more on the total

Why it scored 43: The model card earns 8, with intended use, one limitation and parameters, and no architecture, context, languages or cut-off. Training data earns 5 for a general description of synthetic GUI data with no cut-off. Last change earns 20 for 26 October 2026. Responsiveness earns 0, the one discussion unanswered. Version policy earns 5 for a first release. File health earns 5, as the config still sets an MTP layer the weights lack and the card names a reasoning setting the template refuses.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-upkeep):

- 0 to 20, the model card. Intended use, limitations, architecture, parameters, context length, languages and knowledge cut-off. 20 for all seven, 3 off for each that's missing.
- 0 to 20, training data. 20 for the data published or listed with sources and sizes, 15 for a data card that names the sources or the mix, the cut-off and the filtering, 10 for a general description with a cut-off, 5 without one, 0 for none.
- 0 to 20, the last change to the files, the template or the reference code. 20 within 30 days, 15 within 90, 10 within 180, 5 within a year, 0 older.
- 0 to 20, responsiveness. The publisher's replies to the 40 most recent questions and reports in the repository's discussions or issues. 20 for most, 10 for some, 5 for a few, 0 for none.
- 0 to 10, version policy. 10 for earlier versions kept in their own repositories with a dated note when one is replaced, 5 for earlier versions kept without notes or a first release, 0 when files were replaced in place without a note.
- 0 to 10, file health. 10 when config, tokenizer, template and card agree and no report of a broken file waits without a reply, 5 when one does, 0 when the files contradict the card.

## 4. Agent interface, 70 out of 100, up to 4.6 more on the total

Why it scored 70: The chat template earns 20, Qwen's Jinja file in the repository. The tool-calling format earns 10, as general tool calls rest on the inherited template and the crop action is parsed by inclusionAI's own code, not a runtime parser. Trained tool calling earns 20 for up to two crop actions per turn learned in GRPO. Structured output earns 0. Reasoning control earns 10, with reasoning effort documented and the reasoning shown as think text. Context earns 10 for 262,144 tokens with no scaling.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-interface):

What a model and a developer need to talk to it. Speech, image and other models with no chat interface read the tool-calling and structured-output lines for what the model takes and returns (voices, timestamps, streaming, typed outputs), and the note says how.

- 0 to 20, the chat template. 20 for a template in the repository (Jinja, in the tokenizer config or its own file), 10 when the prompt format is only given as code you run.
- 0 to 25, the tool-calling format. 25 when the publisher documents how tools are declared and how calls come back, with the parser to use in a named runtime, 10 when only the template or a runtime's own recipe says it.
- 0 to 20, tool calling the model was trained for. 20 with several calls in one turn, 10 for one call at a time.
- 0 to 15, structured output. 15 when JSON or JSON Schema output is documented for the model with how to ask for it, 5 when the card claims it without saying how.
- 10, reasoning control. Thinking that can be turned off or budgeted per request, with where the reasoning text comes back.
- 0 to 10, context. 10 for 128K tokens or more trained, 5 for 32K to 127K. Context reached only by scaling the position encoding counts one band down.

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- Capability is computed later. No independent result was found on 27 October 2026 (Artificial Analysis, the ScreenSpot-Pro leaderboard, the PerceptionBench leaderboard, OpenRouter), so Capability stays pending and the total is capped at 61
- GUI grounding (ScreenSpot-Pro, OSWorld-G) isn't among the checklist's areas. The vision area names MMMU, ChartQA, DocVQA and OCRBench, so even a leaderboard row on ScreenSpot-Pro wouldn't count as things stand. Worth deciding whether grounding joins the image area
- File safety 15 takes 10 off for the publisher's scripts that serve the model and write its system prompt. Plain vllm serve loads it with no repository code, and reading only the load path gives 25 and Risk 60
- Tool-calling format 10. Crediting the publisher's documented crop action and shipped parser as the calibration's ship-the-parser reading gives 25 and Interface 85
- Trained tool calling 20 rests on one fixed crop action, two per turn. Reading it as not general tool calling gives 10 or 0
- Reasoning control 10. The publisher's thinking_budget isn't read by the template and its high setting is refused, so a stricter reading gives 0 and Interface 60
- Model card 8 credits the executor line as a limitation. Not crediting it gives 5 and Upkeep 40
- Security programme 5 for AntSRC, as the Ming-3.0 listings read it. Ling-3.0-tiny was given 0 on the same evidence, so the two readings need aligning
- unchecked, GitHub issues on inclusionAI/UI-Venus, since github.com pages weren't fetched (no robots.txt could be read through the session's proxy)
- The Hub's safetensors metadata gives 27,356,728,560 BF16 parameters but a total of 3,054,832; the first figure is used

## Weaknesses

- No independent benchmark result yet; ScreenSpot-Pro, OS-World-G, PerceptionBench and HR-Bench numbers are inclusionAI's own runs
- BF16 only, about 66 GB to load, and no Ollama, LM Studio or quantised build
- The crop loop and system prompt come from the publisher's Python, not a runtime parser or a template
- No safety evaluation, prompt-injection guidance, training data sources or knowledge cut-off
- No hosted API or inference provider serves it

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Serve with `vllm serve inclusionAI/LBS-27B --tensor-parallel-size 4 --max-model-len 65536`, which is what the card's wrapper script runs; keep the context at 32,768 or more for multi-turn images
- Build the system prompt with `build_system_prompt` from the LBS branch's `zoomin_grounding_protocol.py`, send the screenshot after `image_path: <name>`, and don't pass `tools`; the model answers in plain text
- Parse `<action>crop_image(image_path=..., x1=..., y1=..., x2=..., y2=...)</action>`, cut the crop yourself and return it with its mapping to the original image; allow at most two crops per turn and three per episode
- Read the final answer as `[x.xx,y.yy]` on a 0 to 1000 scale of the original image, and use your own executor to click
- Pass `reasoning_effort` as `xhigh`, `medium` or `low` only, since the template raises an error for `high`, and don't count on `thinking_budget`, which the template ignores
- Don't turn on MTP speculative decoding; the MTP weights aren't in the repository though config.json names one MTP layer

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.