Skip to main content

H CompanyLanguage modelOpen weights

Holo4-27B

Holo4-27B has no overall index value yet because there are too few independent benchmark results.

PopularityRank 224 of 342 language models

33out of 100new calculation

based on two data groups, mostly search, range 23 to 42

Summary

Holo4-27B switches between screen interaction, code it writes, and API tools within a task. H Company post-trained Qwen3.8-27B on tasks spanning desktop, web, and Android environments. This dense 27B variant launched alongside Holo4-35B-A3B.

H Company releases the weights under CC BY-NC 4.0 for non-commercial use. The checkpoint handles screenshots and text with up to 262,144 context tokens.

All benchmark results

Every published result with its source, test setup and comparison models.

Show all 17 published benchmark results

Page 1 of 2

Measurements without matching peer values

These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.

OSWorld (Show measurement, test conditions, and source)
Source value
85.2
Metric
OSWorld
Unit
%
Category
source-specific
Direction
Higher is better
Source type
vendor-reported
Evaluator
Provider model card
Status
active
Retrieved at
2026-10-04
Methodology
Vendor-reported result. H Company hai-agents computer-use agent loop with screenshots and tools; the card presents vendor-run results for the exact checkpoint.
Context
Source-specific observation; it is not a shared comparison cohort.
Source Provider model card
OSWorld 2.0 (Show measurement, test conditions, and source)
Source value
61.7
Metric
OSWorld 2.0
Unit
%
Category
source-specific
Direction
Higher is better
Source type
vendor-reported
Evaluator
Provider model card
Status
active
Retrieved at
2026-10-04
Methodology
Vendor-reported result. H Company hai-agents computer-use agent loop with screenshots and tools; the card presents vendor-run results for the exact checkpoint.
Context
Source-specific observation; it is not a shared comparison cohort.
Source Provider model card
AutomationBench 1.0.6 (Show measurement, test conditions, and source)
Source value
45.4
Metric
AutomationBench 1.0.6
Unit
%
Category
source-specific
Direction
Higher is better
Source type
vendor-reported
Evaluator
Provider model card
Status
active
Retrieved at
2026-10-04
Methodology
Vendor-reported result. H Company hai-agents computer-use agent loop with screenshots and tools; the card presents vendor-run results for the exact checkpoint.
Context
Source-specific observation; it is not a shared comparison cohort.
Source Provider model card
OSWorld 2.0 success rate (Show measurement, test conditions, and source)
Source value
41.5
Metric
success rate
Unit
%
Benchmark version
Hcompany Holo4 official release comparison; published 2026-09-28
Direction
Higher is better
Source type
vendor-reported
Evaluator
Hcompany Holo4 release report
Status
active
Retrieved at
2026-10-16
Methodology
Vendor-reported result. Holo4-27B; OSWorld 2.0; success rate. Separate from the previously reported average partial score 61.7%.
Context
OSWorld 2.0 is reported as a single run; full harness details are not stated.
Source Hcompany Holo4 release report
ALE-CLI average score (Show measurement, test conditions, and source)
Source value
44.1
Metric
average score
Unit
%
Benchmark version
Hcompany Holo4 official release comparison; published 2026-09-28
Direction
Higher is better
Source type
vendor-reported
Evaluator
Hcompany Holo4 release report
Status
active
Retrieved at
2026-10-16
Methodology
Vendor-reported result. Holo4-27B; Agents Last Exam CLI 105-task Linux split; Hcompany harness; single run.
Context
Two of 105 attempts that accessed reference answers left on the test machine were excluded; one run; the model's values are vendor-reported.
Source Hcompany Holo4 release report
ALE-CLI success rate (Show measurement, test conditions, and source)
Source value
19.4
Metric
success rate
Unit
%
Benchmark version
Hcompany Holo4 official release comparison; published 2026-09-28
Direction
Higher is better
Source type
vendor-reported
Evaluator
Hcompany Holo4 release report
Status
active
Retrieved at
2026-10-16
Methodology
Vendor-reported result. Holo4-27B; Agents Last Exam CLI 105-task Linux split; Hcompany harness; single run.
Context
Two of 105 attempts that accessed reference answers left on the test machine were excluded; one run; the model's values are vendor-reported.
Source Hcompany Holo4 release report
AndroidWorld (Show measurement, test conditions, and source)
Source value
85.1
Metric
reported score
Unit
%
Benchmark version
Hcompany Holo4 official release comparison; published 2026-09-28
Direction
Higher is better
Source type
vendor-reported
Evaluator
Hcompany Holo4 release report
Status
active
Retrieved at
2026-10-16
Methodology
Vendor-reported result. Holo4-27B; AndroidWorld; score reported by Hcompany as measured by Qwen for Qwen3.8 release.
Context
The source provides no sample count or run variance for this AndroidWorld row.
Source Hcompany Holo4 release report
AutomationBench v1.0.6 held-out 120-task subset (Show measurement, test conditions, and source)
Source value
49.3
Metric
score
Unit
%
Benchmark version
Hcompany Holo4 official release comparison; published 2026-09-28
Direction
Higher is better
Sample
120
Source type
vendor-reported
Evaluator
Hcompany Holo4 release report
Status
active
Retrieved at
2026-10-16
Methodology
Vendor-reported result. Holo4-27B; AutomationBench v1.0.6; held-out subset of 120 tasks; public and held-out results are separate.
Context
Vendor-reported result on 120 held-out tasks within the 600-task public set. The other 480 public tasks belong to the split used to collect training data. This is not the official private-set evaluation.
Source Hcompany Holo4 release report
Agentic Task Factory Web (Show measurement, test conditions, and source)
Source value
80.2
Metric
percent
Unit
%
Direction
Higher is better
Source type
vendor-reported
Evaluator
H Company Holo4 release
Status
active
Retrieved at
2026-10-17
Methodology
Vendor-reported result. H Company's Agentic Task Factory evaluation. Exact Holo4 model column and interface task. The test set mixes in-domain and out-of-domain tasks; the vendor says generated tasks were not used for training.
Context
Vendor-reported. The source publishes no scored sample count or confidence intervals for these tasks. The source does not identify a more specific score metric. The source does not state a benchmark-suite version.
Source H Company Holo4 release
Agentic Task Factory MCP (Show measurement, test conditions, and source)
Source value
89.4
Metric
percent
Unit
%
Direction
Higher is better
Source type
vendor-reported
Evaluator
H Company Holo4 release
Status
active
Retrieved at
2026-10-17
Methodology
Vendor-reported result. H Company's Agentic Task Factory evaluation. Exact Holo4 model column and interface task. The test set mixes in-domain and out-of-domain tasks; the vendor says generated tasks were not used for training.
Context
Vendor-reported. The source publishes no scored sample count or confidence intervals for these tasks. The source does not identify a more specific score metric. The source does not state a benchmark-suite version.
Source H Company Holo4 release
Agentic Task Factory Desktop (Show measurement, test conditions, and source)
Source value
72
Metric
percent
Unit
%
Direction
Higher is better
Source type
vendor-reported
Evaluator
H Company Holo4 release
Status
active
Retrieved at
2026-10-17
Methodology
Vendor-reported result. H Company's Agentic Task Factory evaluation. Exact Holo4 model column and interface task. The test set mixes in-domain and out-of-domain tasks; the vendor says generated tasks were not used for training.
Context
Vendor-reported. The source publishes no scored sample count or confidence intervals for these tasks. The source does not identify a more specific score metric. The source does not state a benchmark-suite version.
Source H Company Holo4 release
PinchBench v2.0.0 (Show measurement, test conditions, and source)
Source value
94.7845804989
Metric
task success rate, score at least 0.5
Unit
%
Benchmark version
2.0.0
Category
agentic-workflow
Direction
Higher is better
Sample
441
Source type
vendor-reported
Evaluator
H Company Trajectories public index
Status
active
Retrieved at
2026-10-17
Methodology
Vendor-reported result. The public index's PinchBench v2.0.0 entry identifies the Holo4 27B series. It describes 147 unique tasks in a staged-file code sandbox, using terminal and file read/write tools, a 200-step cap, and task scores from 0 to 1 with a pass threshold of 0.5. The model entry records 441 task-run trials over three runs and success_rate 0.9478458049886622. The index does not pin a model revision or publish a confidence interval.
Context
Vendor-reported. sampleCount 441 counts repeated task-run trials, not 441 independent tasks: the source identifies 147 unique tasks run three times. No confidence interval or checkpoint revision is published.
Source H Company Trajectories public index

Page 1 of 2

Popularity compared

How often the model is searched for, used and mentioned, compared with every model of its category.

Popularity index

The 100 most popular models of this category, 0 to 100, higher is more popular

20 of 101 models

As of November 2026 · Source: gradually.ai popularity index

Over time

Weekly values from October 18, 2026 to November 20, 2026

Holo4-27BNemotron Cascade 2 30B A3B 33Command A Plus 33Qwen3 Omni 30B A3B Thinking 33Trinity Large Preview 32

Method changes in the timeline, a jump at these points is no market event:

  1. 1November 1, 2026: Value only with search or usage data, 30-day attention
  2. 2November 1, 2026: Method version 4 (common scale, shrinkage)
  3. 3November 9, 2026: New topics measurement V20 (daily quorum, short forms)

What is gaining popularity right now, across models, vendors, harnesses and AI subscriptions, is collected in All AI trends.

All specifications

Architecture

Model class
Vision-language model for computer use
Source: Model class
Architecture
Qwen3.8 dense
Source: Architecture
Parameters
27 billion, dense
Source: Parameters

Input and output

Context window
262,144 tokens
Source: Context window
Input
Text, images, and tool results
Source: Input

Watermarking and provenance

AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.

AI watermarking
Unknown
no source
Watermarking technique
Not known
no source

Licence and access

License
CC BY-NC 4.0
Source: License
Shutdown
None announced
no source

API

Model ID
Hcompany/Holo4-27B
Source: Model ID

Frequently asked questions about Holo4-27B

How large is the context window of Holo4-27B?

262,144 tokens. Separate limits may apply to input and output.

Context windowContext limit in tokens · higher is better
Holo4-27BRank 77 of 112262K
Kimi K2.6262K
Qwen 3.8 27B262K
Kimi K2.7 Code262K
Kimi K2.5262K
Kimi K2 Thinking262K
Mistral Medium 3.5262K

Can I run Holo4-27B locally?

Yes, the weights are open.

Stated by H Company

Open weights
yes
License
CC BY-NC 4.0
Parameters
27B
active per token
n/a

Sources and data date

Show all 10 sources

Benchmark sources at a glance

Which independent evaluators tested this model. This is coverage, not a quality ranking.

No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.

Without vendor claims and historical or disputed results; those stay labelled in the single values.

TypeEvidence and data date
Research statusResearch date October 4, 2026. 3 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Not found in the inspected sources

independent-agentic-task-factory-evaluation

Note

The report's Agentic Task Factory rows are vendor-internal evaluations on task-factory data, not external benchmarks; excluded from this external-benchmark bundle.

Additional sourceHolo4 release (retrieved October 3, 2026) · H Company Holo4 release · Editorial description reviewed October 4, 2026
Additional sourceHolo4-27B model card (retrieved October 3, 2026; October 4, 2026; October 16, 2026) · Holo4-27B model card · Editorial description reviewed October 4, 2026
Additional sourceHcompany Holo4 release report (retrieved October 16, 2026; October 17, 2026; October 22, 2026)
Additional sourceH Company Trajectories public index (retrieved October 17, 2026)
Additional sourceOpenRouter: Popularity index (derived), CC BY 4.0
Additional sourceVercel AI Gateway: Popularity index (derived), CC BY 4.0
Additional sourceOllama: cumulative downloads for the popularity index
Additional sourceReplicate: cumulative runs for the popularity index
Additional sourceGoogle Trending Now: limited feed selection for Germany and the US
Additional sourceCreative Commons Attribution 4.0