Your Startup News Platform - Stay Updated with the Latest!
Startupory
Subscribe
Startup Trends

Krutrim Launches Krutrim-2: India's Largest Open-Weight Multilingual Foundation Model

By , Trends & Analysis Editor6 min read
Share:
Krutrim launches Krutrim-2 70 billion parameter Indian multilingual foundation model

Ola-backed AI lab Krutrim has launched Krutrim-2, a 70 billion parameter open-weight multilingual foundation model trained on 22 Indian languages, marking the largest open-source release from an Indian AI lab.

Key Facts

Key facts: Krutrim Launches Krutrim-2: India's Largest Open-Weight Multilingual Foundation Model
DetailValue
SectorKrutrim
CoverageStartup Trends

Krutrim has released Krutrim-2, a 70-billion parameter open-weight multilingual large language model, available for commercial use under a permissive license.


Model Snapshot


  • Parameters: 70 billion (dense transformer)
  • Languages: 22 Indian languages + English
  • Training tokens: 4.8 trillion (60% Indic, 40% English/code)
  • Context window: 128K tokens
  • License: Krutrim Community License (free for commercial use under 50M MAU)

  • Benchmark Performance


  • IndicGenBench: Beats Llama 3.1 70B by 14 points on average across Indic languages
  • MMLU (English): 79.2 (competitive with Llama 3.1 70B at 79.5)
  • HumanEval (code): 73.8
  • Indic safety eval: Highest score among open models for Indian context

  • Infrastructure


    Trained on Krutrim's in-house Bhumi-1 GPU cluster (16,000 H200 GPUs) over 92 days. The lab also released Krutrim Studio, a hosted inference platform with API pricing 40% below comparable global providers for Indic workloads.


    Why It Matters


    Krutrim-2 is the strongest evidence yet that India can build globally competitive foundation models. The open-weight release benefits the entire ecosystem of Indian AI startups, enterprises and government deployments that need cost-effective Indic-first LLMs.


    What's Next


    Krutrim has committed to releasing a vision-language variant (Krutrim-V) in Q3 and a smaller 8B model optimised for on-device inference by year-end.

    krutrimaifoundation modelllmopen sourceolaindic ai

    Reported by Karthik Iyer. Spotted an error in this story? Request a correction or email corrections@startupory.com. See our editorial policy.

    Share:

    Enjoyed this article?

    Get weekly startup insights delivered to your inbox.