Holo 3.1
Holo 3.1 is H Company's computer-use vision-language model family for agents that inspect screens, reason over UI state, and act across web, desktop, and mobile environments.
Holo 3.1 matters because computer-use agents are moving from browser demos into local, desktop, mobile, and enterprise workflows where function calling, screenshot handling, quantized checkpoints, latency, and API pricing all affect whether an agent can run in production.
H Company announced Holo3.1 on June 1, 2026 as the next generation of its computer-use model family. Official H Company docs and the Models API page confirm the Holo3.1 35B-A3B API model ID, OpenAI-compatible endpoint, text and image input, text output, 65,536 token context, 5-image limit, native function calling, structured outputs, Apache 2.0 license, free-tier rate limit, and token pricing. Hugging Face confirms the Holo3.1 open-weight collection and model sizes, while Reddit discussion is useful only as community demand around local GGUF and self-hosting.
- Build browser, desktop, or mobile computer-use agents that inspect screenshots and choose next actions.
- Use H Company's OpenAI-compatible API for Holo3.1 35B-A3B experiments.
- Evaluate native function calling versus structured-output JSON loops in an agent harness.
- Try local or edge deployments with Holo3.1 open weights and quantized checkpoints.
- Compare hosted API cost, local latency, and privacy tradeoffs for UI automation workloads.
H Company describes Holo3.1 as a family of vision-language models for computer-use agents. The release expands Holo beyond browser and desktop automation into mobile environments, adds native function-calling support for agent frameworks, and ships quantized checkpoints for local inference.
- Release date: June 1, 2026 on H Company's Holo3.1 page.
- Environments: web, desktop, and mobile computer-use workflows.
- Deployment targets: hosted H Company Models API, open weights, and quantized local checkpoints.
The structured model record on GetLLMs uses the H Company Models API entry for holo3-1-35b-a3b, because it has stable catalog fields: model ID, endpoint, input and output modalities, context length, max images, function-calling support, free-tier access, pricing, and license.
Holo3.1 is also relevant to local-agent builders because H Company and Hugging Face list 0.8B, 4B, 9B, and 35B-A3B family sizes, with FP8, NVFP4, and Q4 GGUF checkpoints for local or edge deployment. Treat local performance and hardware fit as deployment-specific until tested on the target machine.
Use H Company docs, the H Company pricing page, and Hugging Face model cards for factual model and API claims. Community posts are useful for search intent, setup friction, and local-running questions, but they should not override official model specs, pricing, license, or benchmark claims.
The runtime loop that feeds screenshots, state, tool results, and actions around a computer-use model.
The provider-facing API fields that make Holo3.1 comparable with other callable models.
Adjacent local-agent workflows where model size, quantization, and hardware fit become practical constraints.
Open-source computer-use agent project that helps readers understand the harness side of screen-control agents.
Agentic RPA product that shows the production infrastructure side of computer-use automation.
Hosted long-context model page useful for contrasting API pricing and agent-loop costs.
Holo 3.1 FAQ
Page-level questions for Holo 3.1.
What is Holo 3.1?+
Holo 3.1 is H Company's computer-use vision-language model family for agents that operate across web, desktop, and mobile environments. The family includes hosted API access through H Company and open-weight checkpoints on Hugging Face.
Which Holo 3.1 model ID should I use with the H Company API?+
Use holo3-1-35b-a3b for the Holo3.1 35B-A3B Models API endpoint. H Company documents the base URL as https://api.hcompany.ai/v1/ and says the API is OpenAI-compatible, so the main client change is the base URL and API key.
Does Holo 3.1 support function calling?+
Yes. H Company documents native function calling for Holo3.1, along with structured-output JSON mode. Structured outputs are useful for constrained agent loops, while native function calling is available on the Holo3.1 API model.
Can Holo 3.1 run locally?+
Yes, depending on the checkpoint and hardware. H Company says the Holo3.1 family includes open weights and quantized checkpoints such as FP8, NVFP4, and Q4 GGUF. Local performance, memory, and latency should still be tested on the actual target machine.