Krutrim-2
India's 12B multilingual model covering English plus 22 Indian languages.
About
Krutrim-2 is Ola's Krutrim AI Lab flagship language model, released February 2025 as a 12-billion-parameter dense transformer built on the Mistral-NeMo-12B-Instruct architecture. It is natively multilingual, generating responses in English and 22 Indian regional languages with a 128K-token context window, and the company reports near-perfect accuracy on Indic benchmarks like IndicXTREME and IN-22 plus strong code generation scores. The weights are on Hugging Face, and Krutrim released companion models alongside it: Chitrarth-1 (multilingual vision-language) and Dhwani-1 (Indic speech recognition).
Limitations: The weights ship under the Krutrim Community License for academic and research use — not a fully permissive open license, so commercial deployment needs legal review. Krutrim's public model momentum has visibly slowed since (the company pivoted toward enterprise cloud), so treat Krutrim-2 as a fixed checkpoint rather than a maintained model line. Independent third-party benchmarks are thin; most performance claims are company-reported.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Company-reported: near-perfect on IndicXTREME and IN-22
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 6/10
Scored 6/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 53; Trending score 62/100 from researcher assessment.