Models API

Holo vision-language models that operate real user interfaces across web, desktop, and mobile: what they do, which size to pick, open weights, pricing.

Models

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. IDs are stable: before a model is removed, its deprecation_date is set in GET /v1/models and a notice appears on this page, so pin an ID in production and check deprecation_date when you upgrade.

What Holo does

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

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.

Smaller sizes for your own hardware

The API serves the two largest models. Three smaller Holo3.1 checkpoints are not on the API, but they are open weights under Apache 2.0: download them and run them yourself. The 35B is open too, with FP8, NVFP4, and GGUF builds for a Mac (M3 or newer, 36 GB) or a DGX Spark.

Local inference covers memory, measured speed, and server setup with llama.cpp or vLLM.

Data retention

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