Models API

Three ways to use Holo

Agent loop

Multi-turn. Conversation and screenshots in, the next action out. Holo as the brain of your own agent.

Element localization

Single-turn. A screenshot and a target description in, click coordinates out.

Document OCR

Single-turn. A page image in, its full text out.

Prefer to try Holo without writing code? HoloTab runs it in your browser with no setup.

Run Holo on your own GPU

The API is the fastest way to start, but you do not need it. Holo3.1 is open weights under Apache 2.0. The Holo3.1 collection on Hugging Face has four sizes, from 0.8B for on-device to the 35B served by this API, plus FP8, GGUF, and NVFP4 checkpoints of the 35B for smaller GPUs.

Model	Best for
Holo3.1 35B-A3B	Same model as `holo3-1-35b-a3b`. Also in FP8, GGUF, and NVFP4.
Holo3.1 9B	Dense model. The most capable you can run on one GPU.
Holo3.1 4B	Smaller and faster on one GPU. Pair it with HoloDesktop CLI to run fully local.
Holo3.1 0.8B	On-device.

See Hosted or local models for llama.cpp and vLLM setups.

Benchmarks

Holo3.1 35B-A3B against the previous generation and comparable open models. Best score per row in bold.

	Holo3.1 35B-A3B	Holo3 35B-A3B	Qwen3.5 397B-A17B	Qwen3.5 122B-A10B	Qwen3.5 35B-A3B	Kimi K2.5	Claude Sonnet 4.6
Overall	78.3	73.3	70.9	67.0	65.9	69.5	67.4
OSWorld	80.0	77.8	62.1	61.6	54.2	63.3*	72.5*
Android World	79.3	67.2	69.0		71.0*
E-Commerce	97.8	94.1	92.2	83.3	80.7	84.4	61.1
Business Software	90.1	86.3	82.4	62.9	69.7	84.7	66.4
Collaboration	75.3	76.0	66.3	60.3	57.7	69.7	52.8
Multi-Apps	65.5	50.0	61.9	47.6	42.9	64.3	69.0
ScreenSpot-Pro	71.5	67.6	69.3	65.5	64.5
OSWorld-G	78.8	77.3	78.4	77.5	76.9

Overall averages the four H Corporate benchmarks (E-Commerce, Business Software, Collaboration, Multi-Apps) first, then takes the mean across OSWorld, Android World, H Corporate, ScreenSpot-Pro, and OSWorld-G. Baselines are evaluated by H; starred numbers are reported from the original source. OSWorld runs on H’s internal implementation of the benchmark, so numbers differ slightly from the official OSWorld-Verified leaderboard.

FAQ

Which model should I use?

Start with holo3-1-35b-a3b. It is on the free tier, has native function calling, and its latency suits interactive loops. Move to holo3-122b-a10b for complex multi-step tasks and novel environments. Switching is a model ID change.

How do rate limits and credits work?

Tier	Rate limit	Models	How to get there
Free	10 requests per minute	`holo3-1-35b-a3b`	Create an account. No credit card.
Paid	Higher limits	Both models	Add credits. Up to 15 minutes to appear.

The credit dashboard shows your balance and usage.

Can my balance go negative?

By a few cents, yes. The balance is checked before a request runs, so a large request on a low balance can overshoot. If it keeps growing, contact support@hcompany.ai.

Do credits expire? Are they refundable?

Unused credits may expire one year after purchase and are not refundable by default. Regional obligations may apply. See the terms of use.

What data is logged?

Request time, model, and token counts. Prompts and responses are not stored: the API uses zero data retention by default. H runs its own models, so inputs and outputs are never shared with third parties. See the privacy policy.

Do model IDs change?

No. IDs are stable. Before a model is removed, its deprecation_date is set in GET /v1/models and a notice appears on this page. After removal, requests to the old ID fail with model_not_found. Pin an ID in production and check deprecation_date when you upgrade.

Get started

Guides

Models

Three ways to use Holo

Agent loop

Element localization

Document OCR

Run Holo on your own GPU

Benchmarks

FAQ

​Models

​Three ways to use Holo

Agent loop

Element localization

Document OCR

​Run Holo on your own GPU

​Benchmarks

​FAQ

Models

Three ways to use Holo

Run Holo on your own GPU

Benchmarks

FAQ