Ola chief Bhavish Aggarwal on Tuesday introduced that Krutrim Cloud will be capable of run Meta’s Llama 4 fashions totally on an India-hosted cloud infrastructure. The transfer will permit builders throughout the nation to entry superior AI capabilities whereas sustaining full information sovereignty.
“Excited to share that Krutrim is among the many world’s first to host Meta’s Llama 4 fashions working totally on its India-hosted cloud. Powering our builders with world-class AI, at industry-disrupting costs with full information sovereignty,” he mentioned in a put up on X.
In a separate LinkedIn put up, he mentioned that the corporate is deploying each Llama 4 Scout and Llama 4 Maverick fashions at much more disruptive costs – simply ₹7 to ₹17 per million tokens.“This isn’t nearly value financial savings – it’s about democratising entry to cutting-edge AI for each Indian developer and startup,” he mentioned
Llama 4 fashions, together with Scout and Maverick, at the moment are dwell on its platform, permitting builders to construct and deploy AI purposes at aggressive pricing. The fashions are hosted inside India’s borders, aligning with rising calls for for localised information management and privateness.
Krutrim Cloud, launched final 12 months, supplies a complete suite of AI companies, together with Mannequin-as-a-Service (MaaS) and GPU-as-a-Service. It lately added assist for DeepSeek fashions as effectively.
Meta lately launched two multimodal open-weight fashions—Llama 4 Scout and Llama 4 Maverick. Each fashions are constructed on a mixture-of-experts (MoE) setup.
Llama 4 Scout options 17 billion energetic parameters and 16 specialists, designed to suit inside a single H100 GPU. Meta claims it helps an industry-leading 10 million token context window, enabling advanced duties comparable to multi-document summarisation and reasoning over giant codebases.
Llama 4 Maverick is a 17 billion energetic parameter mannequin with 128 specialists. It contains 400 billion complete parameters and performs competitively with bigger fashions like DeepSeek V3 on reasoning and coding duties. Meta mentioned that Maverick exceeds GPT-4o and Gemini 2.0 Flash on a number of benchmarks. It scored an ELO of 1417 on LMArena in experimental chat settings.
The fashions had been distilled from Llama 4 Behemoth. This unreleased instructor mannequin can also be a multimodal mixture-of-experts mannequin, with 288B energetic parameters, 16 specialists, and practically two trillion complete parameters.
There have been additionally some questions across the coaching and testing information of the mannequin, which had been later clarified by Ahmad Al-Dahle, the lead of GenAI at Meta. “That’s merely not true, and we’d by no means do this. Our greatest understanding is that the variable high quality persons are seeing is because of needing to stabilise implementations.”
The put up Meta’s Llama 4 Fashions Now Accessible on Krutrim Cloud appeared first on Analytics India Journal.
