Holo4-35B-A3B

Released
2026-09-28
Last refreshed
2026-11-14
Status
Researched today
Open sourceCommercial use: permittedMultimodalRAGAgentsLong contextVisionJSON / Tool useAgentsVision
H Company releases · 11 in the last 12 months · this family litChangelog →
Specifications
Family
Holo
Released
2026-09-28
Context
256K
Max output
16K
Parameters
35B (3B active)
Architecture
Mixture of Experts
Specialization
computer-use
Openness
Open source
License
Apache 2.0OSI-approvedPermitted
Weights
Available
Code
Unknown
Training
Reinforcement Learning
Created by

Agentic AI for Action

Paris, France
Founded 2023
Website
Pricing
Output / 1M
$2.00
Input / 1M
$0.300

Cheapest of 1 route · H Company API · cache read $0.030

Providers(1)

About

Holo4-35B-A3B is H Company's 35B-parameter (3B active) Mixture-of-Experts vision-language model for computer-use agents, released September 28, 2026 alongside Holo4-27B. Built on Qwen3.6-35B-A3B, it reads screenshots and text and acts by clicking and typing, writing and running its own code, or calling MCP and API tools, across web, desktop and Android. It has a 262,144-token context, thinking that can be switched off, and native function calling. Open weights in BF16, FP8, NVFP4 and GGUF are released under Apache 2.0, and it is served on the H Models API (holo4-35b-a3b).

Providers(1)

Compare API pricing across 1 tracked provider route for input and output tokens, batch, and cached reads when available.

ProviderInput / 1MOutput / 1MCacheRoute
H Company API$0.300$2.00read $0.030
Serverless

Capabilities

VisionMultimodalReasoningJSON / Tool usePrompt Caching

Benchmark peer barsfor RAG

No task-mapped benchmark peers are available for this model yet.

Benchmark scores(1)

Scores are benchmark-specific and are direction-aware: the same numeric gap can mean very different outcomes across suites. Use the leaderboard context and this model's provider route to decide whether the winning margin is meaningful for your workload.
BenchmarkScoreVersionEvaluationSource
77.6AndroidWorld - mobile agentObserved 2026-09-28
Configuration: Holo4-35B-A3B (H Company harness)
Harness: H Company agent harness (mean over 2 to 4 runs)
Evaluator: H Company
Confidence: self-reported
Notes: H Company self-reported on the Holo4 launch post (benchmark table, AndroidWorld 'Phone apps' 77.6%, $0.07 per task at H Models API rates), retrieved 2026-11-14. Footnote: 'Holo4 in our harness: mean over 2 to 4 runs'.

Migration checks

No linked migration route is available for this model yet.

API versions

holo4-35b-a3bHcompany/Holo4-35B-A3BHcompany/Holo4-35B-A3B-FP8Hcompany/Holo4-35B-A3B-NVFP4Hcompany/Holo4-35B-A3B-GGUF