ZenteiQ logo

ZenteiQ models

ZenteiQ is an IISc Bangalore spin-out that builds the Dhīr compact foundation models, pretrained from scratch in India, alongside scientific machine learning tools.

Bengaluru, Indiazenteiq.ai

Releases
2
Open weights
2
Latest
Nov 03, 2026

November 2026

2 releases
  1. Small modelOpen weights

    Dhīr 2B

    Dhīr 2B is ZenteiQ's dense 2.03B-parameter model, pretrained from scratch in India for English and ten Indian languages and sized to run offline on laptops and edge devices.

    Modalities
    text → text
    Context
    16K context
    Parameters
    2.03B
    License
    Apache-2.0
    At launch
    Open weights, Enterprise
    • Pretrained on more than 6 trillion tokens on Google Cloud TPU v6e in India, with an in-house 131k-entry tokenizer
    • Instruction tuned with SFT and GRPO, with native tool calling that ZenteiQ scored at 72.4 on BFCL v3
    • ZenteiQ reported first place in its class on 13 of 15 Indic benchmarks
    • A 4-bit file of about 1.21 GB ran at 18.6 tokens per second on Intel Iris Xe graphics in ZenteiQ's test
    Source: Dhīr 2B and Dhīr 5B MoE | ZenteiQ
  2. LLMOpen weights

    Dhīr 5B MoE

    Dhīr 5B MoE is ZenteiQ's sparse mixture-of-experts model, pretrained from scratch in India for English and ten Indian languages and built for private, offline enterprise assistants.

    Modalities
    text → text
    Context
    16K context
    Parameters
    MoE: 5.09B total, 1.79B active
    License
    Apache-2.0
    At launch
    Open weights, Enterprise
    • 12 routed experts per layer with top-2 routing and no shared experts
    • Pretrained on more than 6 trillion tokens on Google Cloud TPU v6e in India, with an in-house 131k-entry tokenizer
    • ZenteiQ reported a BFCL v3 tool-calling score of 76.3 and higher agentic, code and math averages than Dhīr 2B
    • A 4-bit file of about 2.95 GB runs offline; ZenteiQ said the model is text only and has no reasoning mode
    Source: Dhīr 2B and Dhīr 5B MoE | ZenteiQ

Cite and reuse this data

These release records are free to reuse under CC BY 4.0, including commercially. Credit The AI Engineering and link to this page's URL.

Dates are the labs' official announcement dates, and every record links to its source. List fields in the CSV are separated by " | ".

Citation

The AI Engineering Editors. "ZenteiQ models: release history and timeline." The AI Engineering, updated Nov 03, 2026. https://theaiengineering.com/releases/zenteiq

HTML link

<a href="https://theaiengineering.com/releases/zenteiq">ZenteiQ models: release history and timeline</a> (The AI Engineering)
Full dataset
1,250 releases from 73 labs
Coverage
to
Updated
License
CC BY 4.0
Publisher
The AI Engineering

Other labs