Ultineo Reasoning

The models that run your reasoning

A library of open models, from the most compact to the most thorough. Each application picks its own, and each step of a chain keeps the model of its application.

Talk to us about Ultineo ReasoningSee the plans

A model for every task

Your choice

Each application has its own model. Change it at any time, without touching its fields or its instructions.

Open models

The library brings together open-weight models published by recognised makers.

Nothing to install

You pick the model; Ultineo runs it and measures the usage. There is no provider key to manage.

The library

The model families on offer, with their sizes in billions of parameters and what they can read.

  • Qwen

    Alibaba

    General purpose, with vision models

    Qwen3.8, Qwen3.5, Qwen3, Qwen3 VL

    4 to 27 billion parametersText and image

  • Gemma

    Google

    General purpose, translation, code and content safety

    Gemma 4, Gemma 3, Translate Gemma, Code Gemma, Shield Gemma

    2 to 27 billion parametersText and image

  • Mistral

    Mistral AI

    General purpose, code and mathematics

    Mistral Small 3.2, Ministral 3, Devstral Small 2, Codestral, Mathstral, Mistral

    3 to 24 billion parametersText and image

  • Granite

    IBM

    Business tasks, code and content safety

    Granite 4.1, Granite 4, Granite 3.3, Granite 3.2 Vision, Granite Code, Granite Guardian 4.1

    2 to 8 billion parametersText and image

  • Phi

    Microsoft

    Compact, with step-by-step reasoning variants

    Phi 4, Phi 4 reasoning, Phi 4 mini, Phi 4 mini reasoning, Phi 3.5

    4 to 14 billion parametersText

  • DeepSeek

    DeepSeek

    Step-by-step reasoning, code and document reading

    DeepSeek R1, DeepSeek V2, DeepSeek Coder V2, DeepSeek OCR

    2 to 16 billion parametersText and image

  • MiniCPM

    OpenBMB

    Compact vision: images and documents

    MiniCPM-V 4.6, MiniCPM-V 4.5

    1 to 8 billion parametersText and image

  • LFM

    Liquid AI

    Fast and efficient

    LFM2.5

    8 billion parametersText

The library evolves: new models join it regularly, and the largest sizes depend on your plan.

Choosing by task

Write and summarise

A mid-sized general-purpose model is enough for most texts.

Sort and classify

A small model answers fast: ideal for a sorting step in a chain.

Read an image

A vision model reads a photo, a screenshot or a scanned document.

Translate

A dedicated translation model keeps meaning and tone from one language to another.

Write code

Code models generate, explain and fix programs.

Reason step by step

Reasoning variants lay out their thinking before concluding.

Combine models in a chain

Not sure which model to pick?

Describe your task and its volume; we will point you to the right model and size.

Talk to us about Ultineo Reasoning

Questions about Ultineo Reasoning

Yes, at any time. Its fields and instructions stay in place.

No. Each plan gives access to a range of model sizes; the largest are reserved for paid plans.

No. Your fields, instructions and answers stay in your project and are not used to train any model.

No. The library’s models are run by Ultineo; there is no third-party key to obtain.