
Institute · Abu Dhabi, United Arab Emirates
Institute of Foundation ModelsLLM360
MBZUAI research institute that sponsors LLM360, the fully open effort behind the Amber, K2 and TxT360 releases, and the K2 Horizon model family.
Abu Dhabi, United Arab Emirates
- Type
- Institute
- Based in
- Abu Dhabi, United Arab Emirates
- Key papers
- 2
- Featured projects
- 4
About
The Institute of Foundation Models (IFM) is a research unit of the Mohamed bin Zayed University of Artificial Intelligence in Abu Dhabi, launched publicly in May 2025. The LLM360 project, which states it is proudly sponsored by the institute, publishes open large language models together with their training data, code and checkpoints. The LLM360 Hugging Face organization notes that its namespace has moved to IFM.
LLM360's releases include Amber-7B, Crystal-7B, the 65-billion-parameter K2 trained on 1.4 trillion tokens, and TxT360, a pre-training dataset that deduplicates 99 Common Crawl snapshots and 14 other sources. In September 2026 the institute released K2 Horizon, six fully open models from 0.9 to 375 billion parameters.
The institute also coordinates the Jais Arabic models and operates with a Silicon Valley lab alongside its Abu Dhabi base.
Mission and ambitionsLLM360 aims to enable community-owned AI through open large model research, creating standards and tools for capability and knowledge transfer. The institute releases weights, code, data and training details.
Known for
- 01
LLM360 fully open model releases
- 02
K2 and K2 Horizon models
- 03
TxT360 pre-training dataset
Milestones
- 2023
Amber-7B and Crystal-7B
Released open 7B models with full training artifacts and chat variants.
- 2024
K2-65B
Released a 65B model trained on 1.4 trillion tokens that LLM360 reports matches Llama 2 70B with less compute.
- 2024
TxT360
Released a pre-training dataset globally deduplicating 99 Common Crawl snapshots and 14 other sources.
- 2025
Institute launch
MBZUAI launched the Institute of Foundation Models and a Silicon Valley lab.
- 2026
K2 Horizon
Released six fully open models from 0.9B to 375B parameters with data and methodology.
Open source
Pretraining code for Amber 7B.
Open research on large reasoning models.
Open LLM analysis implementations.
Open 370B-token math pretraining dataset.
Research papers
Openness
LLM360 is known for releasing training artefacts; the legacy HF org shows no public models.
People
- Eric XingPresident of MBZUAI; Professor, Carnegie Mellon University
Leads the university that hosts the Institute of Foundation Models and commented on K2 Think.
Listed from the lab's own pages and public profiles.
Contact and community
llm360.ai lists Twitter, Discord, GitHub and email links.
Discord linked from llm360.ai for the LLM360 community.
Sources
Compiled from the lab's public pages, Hugging Face and GitHub. Last checked 2026-11-17. Something wrong or out of date? Tell us on Discord.



