Home / Models / Holo4-27B

H COMPANY · FIELD VERIFIED

Holo4-27B

Primary sources last checked on 12 Oct 2026. Verification freshness is tracked separately from factual change, and unknown values remain explicit rather than inferred.

NotableNotable for a newly released open-weight computer-use VLM with official model weights, first-party GUI/tool-harness documentation and publicly released agent trajectories. H Company's OSWorld 2.0 figures are publisher-reported, not independent OMW agent benchmark results. CC BY-NC 4.0 prohibits ordinary commercial use without separate rights.Curation method →

Source repository ↗Compare this model →History & diff →Machine-readable JSON ↗Field-verified 2026-10-12

Evidence statusSource-first verified

Last verified12 Oct 2026

Full field verification12 Oct 2026

Source policy →Methodology →

FRESHNESS & VERIFICATION

Current evidence, dated honestly.

A verification check does not pretend the model changed. Open Model Weights updates the verification date after a real primary-source check, while factual-update dates move only when retained source evidence changes.

Last verified: 12 Oct 2026
Last factual update: No change observed
Repository created: 24 Sep 2026

View observed history →View evidence JSON →View model JSON →

CONFLICT INTELLIGENCE

Claims stay attached to evidence class.

No active comparable-source conflict detectedThis is not a claim that every possible source agrees; it covers the comparable fields currently checked.

Publisher claims3

Source observed4

OMW derived3

OMW measured0

Conflict Intelligence →Conflict JSON ↗Evidence IDs ↗OMW measurement policy →

+ Save to my shortlist Monitor changes →Local browser monitoring · no background email alerts

DIRECT ANSWER · FEATURED MODEL BRIEFING

Holo4-27B in plain English.

Source-grounded editorial layer · reviewed 2026-10-12

Holo4-27B is H Company's dense vision-language checkpoint for computer-use agents. The official Hugging Face repository publishes downloadable weights and identifies Qwen3.8-27B as its base. H Company documents GUI actions, tool calls and code execution through its separate agent harness. The checkpoint is licensed CC BY-NC 4.0: commercial usage rights must not be inferred from the Apache license of its base model. Published performance figures are vendor results, not Open Model Weights measurements.

WHY IT MATTERS

A GUI-oriented model can be selected separately from the orchestration and execution harness that makes an agent act on a desktop.

The released vendor trajectories offer more inspection context than an unsupported 'agent-ready' label.

The non-commercial license is a material constraint for business deployment, even when the base model used a permissive license.

MODEL-SELECTION VIEW

Strong fit when…

Non-commercial research into computer-use models with public first-party evaluation traces.
Controlled experiments on GUI grounding, code execution and tool selection under a sandboxed harness.

Important considerations

CC BY-NC 4.0 governs the distributed Holo4 weights; do not assume business usage is permitted.
An OSWorld 2.0 number published by the model developer is not an independent or cross-version comparison.
GUI agents require real isolation, permissions, action logs, and human approval for sensitive steps.

Model architecture and task boundaries

H Company identifies Holo4-27B as a dense 27B vision-language model adapted from Qwen3.8-27B. The model reads screen imagery and proposes actions, but execution requires a separate controller.

The 262,144-token context length is publisher-declared and must be checked against the source configuration. It is not evidence of accurate long-horizon agent behavior.

Reported GUI evaluation and provenance

The publisher reports 61.7% on OSWorld 2.0 using its Holo4 setup and releases agent trajectories. These results must be attributed to H Company; no independently reproduced Open Model Weights run exists.

OSWorld 2.0 and other computer-use benchmarks have different task releases and agent harnesses. Scores from different protocol settings should never be pooled.

Licensing and commercial restrictions

The model card states the Holo4 weights use Creative Commons Attribution-NonCommercial 4.0 even though the base Qwen model is published under Apache 2.0. License rights attach to the distributed checkpoint, not merely to its base model.

A commercial deployment requires explicit applicable rights and technical safeguards. This page does not provide a legal opinion.

Practical evaluation before local deployment

Start with the official model artifact and H Company's agent harness documentation. Record the exact weights revision, chat template, screenshot preprocessing, tool interface and execution environment.

Use a disposable desktop and non-sensitive tasks. Test coordinate grounding, incorrect-tool recovery, permission boundaries and repeatability before considering any real-user workflow.

PRIMARY & INDEPENDENT SOURCES

H Company official weights and licensingInspect ↗H Company Holo4 release and OSWorld methodInspect ↗

FAQ

Questions people ask about Holo4-27B.

Can I commercially deploy Holo4-27B?

The official model-card terms identify CC BY-NC 4.0 for these weights; ordinary commercial use is not permitted by that non-commercial license. Check the original terms and any separately granted rights.

Does Holo4-27B include a complete computer-use agent?

No. The checkpoint must be paired with an orchestration harness, GUI or API tools, permissions, logging and execution controls.

MODEL INTELLIGENCE · SOURCE-FIRST

What is Holo4-27B?

OMW evidence review12 Oct 2026

Open Model Weights records Holo4-27B as a 27.3567B-parameter text + image → text/actions open-weight model in the Holo4 family, with H Company listed as the developer. The checked source repository provides 19 recognized weight artifacts, format evidence for Safetensors and a verified context window of 262,144 tokens. Values that are not evidenced in the checked sources remain explicitly undisclosed or unknown.

The repository declares cc-by-nc-4.0. Commercial use not allowed by the declared non-commercial license. Open Model Weights reports the checked license evidence and classification; this is not legal advice.

The checked sources declare Qwen/Qwen3.8-27B as lineage evidence. Training recipe status: training-described-in-model-card; data disclosure status: described-in-model-card.

Source repository ↗Model card ↗Configuration ↗License evidence ↗

DEPLOYMENT PROFILE

Weight memory, without pretending it is a deployment guarantee.

These bars visualize weight-only decimal-GB estimates. They exclude KV cache, activations, runtime overhead, optimizer state and sharding.

WEIGHT-ONLY MEMORY

BF16 / FP16~54.7 GB

FP8 / INT8~27.4 GB

INT4~13.7 GB

FIELD EVIDENCE MAP

WeightsVerified

LicenseVerified

Commercial useNot allowed

ContextVerified

FormatsVerified

LineageDeclared

Training recipeTraining described in model card

Training dataDescribed in model card

RuntimeSource-derived · not runtime-tested

LICENSE & USE

cc-by-nc-4.0

Commercial use not allowed by the declared non-commercial license

Declared licensecc-by-nc-4.0

ClassificationNot allowed

VerificationLicense file checked

This section summarizes the checked repository evidence. It is not legal advice. Inspect license evidence ↗

LINEAGE & PROVENANCE

What the checked sources disclose — and what they do not.

Open Model Weights does not infer a hidden base model or training corpus when the checked sources do not declare one.

LINEAGE

BASE / PARENTQwen/Qwen3.8-27BDeclared

base_model

CURRENT RECORDHolo4-27BHolo4

TRAINING DISCLOSURE

Training recipeTraining described in model card

Training dataDescribed in model card

FamilyHolo4

RUNTIME SIGNALS

Compatibility claims stay attached to their evidence level.

Declared and model-card-mentioned runtimes are discovery signals. They are not treated as Open Model Weights runtime tests unless explicitly measured.

Declaredtransformers

MentionedNeMo

MentionedTransformers

OMW runtime testNot performed

EVIDENCE TRAIL

Sources and observed history.

This is where the article remains auditable: source links on one side, the registry's observed change history on the other.

PrimaryModel Repositoryhttps://huggingface.co/Hcompany/Holo4-27BInspect source ↗PrimaryModel Apihttps://huggingface.co/api/models/Hcompany/Holo4-27BInspect source ↗PrimaryModel Cardhttps://huggingface.co/Hcompany/Holo4-27B/blob/main/README.mdInspect source ↗PrimaryConfighttps://huggingface.co/Hcompany/Holo4-27B/blob/main/config.jsonInspect source ↗Primary-or-declaredLicensehttps://huggingface.co/Hcompany/Holo4-27B/blob/main/LICENSEInspect source ↗

2026-10-12

Field-by-field verification

Checked Hugging Face API metadata, exact repository file list, model card and config when accessible.

01 · WEIGHTS

Exact repository artifacts

Weightsverified-available

Accesspublic

RepositoryHcompany/Holo4-27B

Recognized weight files19

Show all 19 exact weight files

HF API file listing ↗

02 · LICENSE

License & commercial use

Declared licensecc-by-nc-4.0

License IDcc-by-nc-4.0

Commercial useCommercial use not allowed by the declared non-commercial license

Repository license filesLICENSE

Classification describes the checked license terms; it is not legal advice.

License / model-card evidence ↗

03 · HARDWARE

Weight-memory estimate

BF16 / FP16~54.7 GB

FP8 / INT8~27.4 GB

INT4~13.7 GB

Weight-only decimal-GB estimate from total parameter count; excludes KV cache, activations, runtime overhead, optimizer state and sharding.

04 · FORMATS

Formats & precision evidence

Safetensors

BF16: not observedFP16: not observedFP8: not observedINT8: not observedINT4: not observedGGUF: not observed

“Not observed” means no matching repository artifact/name/dtype signal was found; it does not prove no third-party conversion exists.

05 · LINEAGE

Base model & family

FamilyHolo4

Variant27B

Declared base modelQwen/Qwen3.8-27B

Relationbase_model

06 · TRAINING ASSETS

Recipe & data disclosure

Training recipetraining-described-in-model-card

Data disclosuredescribed-in-model-card

EvaluationsNot classified by automated verifier

Intermediate checkpointsNot classified by automated verifier

07 · RUNTIME SUPPORT

Declared, mentioned and externally evidenced runtimes

Declared librarytransformers

Declared/taggedtransformers

Mentioned in model cardNeMo, Transformers

Artifact signalsNone

External primary runtime evidenceNone

Tested by Open Model WeightsNo — source-derived support only

Open Model Weights separates model-repository evidence from external runtime-release evidence. An external support statement does not change the official model repository's weight formats and does not imply an Open Model Weights inference benchmark.

08 · SOURCES

Field evidence

Primary / declared sources5

Repository revision4de03cfeee1d

Source policyHow evidence is classified →

model_repositoryhttps://huggingface.co/Hcompany/Holo4-27B

model_apihttps://huggingface.co/api/models/Hcompany/Holo4-27B

model_cardhttps://huggingface.co/Hcompany/Holo4-27B/blob/main/README.md

confighttps://huggingface.co/Hcompany/Holo4-27B/blob/main/config.json

licensehttps://huggingface.co/Hcompany/Holo4-27B/blob/main/LICENSE

09 · VERIFICATION HISTORY

Checked, not assumed

2026-10-12

Field-by-field verification

Checked Hugging Face API metadata, exact repository file list, model card and config when accessible.

licenseverified

commercial useclassified

contextverified

weight filesverified

formatsverified

base modeldeclared

training recipetraining-described-in-model-card

data disclosuredescribed-in-model-card

runtime supportsource-derived-not-runtime-tested

MODELS FOR AI AGENTS · ROLE EVIDENCE

Where this model fits in an agent system

Computer-use vision-language model. A candidate for a reasoning or coordination role. No end-to-end agent performance is inferred from its model category.

The curated notable tier is not an agent-readiness rating. Model-specific tool-call success, long-horizon task accuracy and compatible runtime configurations are not independently established here.

Source-linked license: cc-by-nc-4.0 (not-allowed commercial-use classification; see original terms).

Context: 262144 configured tokens.

BF16 weights: 54.71 estimated decimal GB, not runtime VRAM.

Compare agent model candidates → · Agent evaluation method → · Inspect original model repository ↗

Agent-specific published evidence

These records are separately sourced. They do not replace independent model-plus-runtime tests.

Computer use · publisher-documented — H Company documents screenshot-guided GUI interaction, code execution and API/MCP tool use via its hai-agents harness. Original source ↗
Tool calling · publisher-documented — The Holo4-27B publisher documents function calling and tool-driven desktop/browser workflows; exact test success is not independently measured by OMW. Original source ↗
OSWorld 2.0 · 61.7 percent — reported by H Company; not OMW-tested. Original source ↗
vLLM · publisher setup documented — Hugging Face displays a model serve example. This does not confirm function-call parsing or complete computer-use harness setup; OMW has not run it. Original source ↗

Read evidence details and benchmarks → · Use the agent model selector →