# Aleph Alpha — AI Compare Ninja

> Language models and the PhariaAI platform for domain knowledge, document work and AI applications in enterprises and public bodies.

## Overview
- **Category:** Productivity (Assistants & workspaces)
- **Status:** Active · enterprise only - Enterprise and public-sector offering with access and deployment agreed individually.
- **Paid from:** contact sales / custom contract (PhariaAI)
- **Free access:** No free plan published
- **Website:** https://aleph-alpha.com/
- **Canonical URL:** https://www.aicompare.ninja/en/tool/aleph-alpha

## What it does
- Kolibri-1: open-weight model (Apache 2.0) with 78.1 billion parameters, 3.46 billion active, and context up to 1M tokens
- Specialized language models and integration of additional models
- PhariaAssistant, PhariaStudio, PhariaOS and PhariaCatch
- On-premises, Kubernetes cloud and hybrid deployment
- Development, evaluation and operations for organization-specific tasks

## In brief
Aleph Alpha develops specialized language models with enterprises and public bodies. The PhariaAI platform also supports open and proprietary models from other providers. PhariaAssistant handles document work, PhariaStudio development and evaluation, PhariaOS operations and PhariaCatch structured expert feedback.

Deployment can be on-premises, in a Kubernetes cloud or hybrid. The documentation lists no general SaaS offering for the PhariaAI stack; managed instances and industry-specific SaaS solutions require separate discussion. Prices and capacity are agreed by contract.

With Kolibri-1, Aleph Alpha released an open language model in October 2026: an English-German mixture-of-experts model with 78.1 billion parameters, of which 3.46 billion are active per token. The weights are on Hugging Face under Apache 2.0 and can run on your own hardware. Training used contexts up to 256k tokens; according to the tech report, up to 1M tokens are usable, with quality that degrades depending on the task. The model card lists reasoning levels (low, medium, high) and tool calling. The FP8 weights (about 78 GB) need at least roughly 2× A100 80 GB or 1× H200. The model card names no hosted API and no price.


## Limits
PhariaAI is not limited to on-premises: the documentation also lists Kubernetes clouds such as StackIT, AWS, Google Cloud and Azure, plus hybrid installations. Air-gapped operation and managed instances need vendor coordination. Model choice, context window, throughput, infrastructure and data processing depend on the installation and agreement; general token allowances or fixed public package prices are not verified. Website privacy statements do not establish a universal training policy for every customer deployment. Kolibri-1 is an open-weight model for self-hosting: the Apache 2.0 licence charges no fee, while hardware and operation are your own cost. The model card and tech report name no hosted API and no token price. The sources checked do not establish whether Kolibri-1 is part of PhariaAI installations.

## Pricing tiers
| Tier | Price | Period | What you get |
| --- | --- | --- | --- |
| PhariaAI | contact sales | custom contract | PhariaAssistant, PhariaStudio, PhariaOS and PhariaCatch; On-premises, Kubernetes cloud and hybrid deployment; Enterprise and public-sector offering with access and deployment agreed individually. |

## Sources
All details regarding plans, limits, and features are based on official vendor documentation and the primary sources verified below.

- https://aleph-alpha.com/
- https://docs.aleph-alpha.com/phariaai-home/latest/index.html
- https://docs.aleph-alpha.com/phariaai-install-config-guide/latest/installation/before-you-start/deployment-options.html
- https://aleph-alpha.com/impressum/
- https://huggingface.co/Aleph-Alpha/Kolibri-1
- https://aleph-alpha.com/downloads/tech-report.pdf

---
*Source: AI Compare Ninja (2026) - https://www.aicompare.ninja/en/tool/aleph-alpha*