Bud Ecosystem built Hex-1 — a 4-billion-parameter open-source LLM tailored for Indic languages — in collaboration with AMD. Trained on AMD Instinct™ MI300X GPUs and the ROCm™ open-source stack, it delivers state-of-the-art performance in Hindi, Telugu, Tamil, Kannada, and Malayalam.
Bud Ecosystem has developed Hex-1, India’s first commercially usable open-source large language model tailored for Indic languages, in collaboration with AMD. Trained on AMD-powered compute infrastructure, Hex-1 bridges the linguistic AI gap for millions of users in India while enabling developers and businesses to build inclusive language applications.
India’s linguistic diversity is both cultural wealth and a challenge for AI. With only about 10% of the population fluent in English, most global AI models are ineffective for the rest of the country’s users. Bud identified this gap and set out to build an LLM that could serve India on its own terms.
Bud set three requirements for the model: that it understand and generate multiple Indian languages reliably; that it be fully open source and commercially usable; and that it enable fast innovation across education, enterprise productivity, customer support, and more.
Releasing Hex-1 under an open-source and commercial license lets startups, developers, and enterprises integrate and innovate freely — without restrictive usage terms — accelerating AI adoption across India.
Hex-1 is a 4-billion-parameter LLM engineered specifically for Indic languages. Unlike many global models trained primarily on English data, Hex-1 was built to handle the linguistic richness and nuance of Indian languages — making generative AI practical, usable, and inclusive for the Indian market.
When benchmarked against leading models like Gemma-2B, LLaMA-3.2-3B, and Sarvam-1, Hex-1 delivers best-in-class performance across Indic languages on key evaluation metrics such as MMLU (Massive Multitask Language Understanding).
Crucially, Hex-1 is released under an open-source and commercial license, enabling startups, developers, and enterprises to integrate and innovate freely without restrictive usage terms — accelerating AI adoption across India.
Bud’s engineering and research teams chose AMD Instinct™ MI300X GPUs — deployed via Microsoft Azure — and the AMD ROCm™ open-source software stack to train Hex-1. This selection was central to meeting Bud’s aggressive training timeline and performance goals. Four advantages carried the work.
AMD Instinct MI300X GPUs helped Bud train Hex-1 up to 30% faster than alternative configurations, shortening development cycles and enabling rapid iteration.
Up to 30% fasterHigh memory capacity and computational efficiency meant Bud used about 40% fewer GPUs, cutting infrastructure costs significantly.
~40% fewer GPUsWith 192 GB of high-bandwidth memory, Bud could run larger batch sizes and longer context lengths without complicated memory workarounds.
192 GB HBMAMD’s ROCm stack, compatible with PyTorch and other open tools, fit Bud’s open-source philosophy and helped streamline model training and optimization.
ROCm™ + PyTorchThis AMD-powered infrastructure gave Bud the compute performance and flexibility to deliver an LLM capable of competing with larger global models — while keeping costs and development time under control.
Best-in-class language understanding, commercial openness, and momentum for India’s AI ecosystem.
Hex-1 isn’t just another language model. Evaluated across Indic language benchmarks for comprehension and reasoning, it frequently outperformed larger models — and its open license puts that capability in every developer’s hands.
When evaluated across Indic language benchmarks for comprehension and reasoning, Hex-1 frequently outperformed larger models, including popular global alternatives. Its strong performance on MMLU and related metrics demonstrates real technical capability for diverse language use cases.
By offering Hex-1 under an open-source license that allows commercial use, Bud enables a broader ecosystem of innovation — from AI-powered learning tools to localized enterprise assistants that truly understand and respond in native Indian languages.
Hex-1 fosters a stronger Indian AI ecosystem by empowering developers, researchers, and companies with a model that’s both high-performing and freely usable — a combination that lowers barriers to adoption and fuels AI-driven solutions tailored to local needs.
With Hex-1, Bud Ecosystem has built a powerful, accessible, and India-centric generative AI model that broadens the reach of language technology. AMD’s high-performance GPUs and open-aligned software played a meaningful role in enabling its training and optimization — helping Bud deliver superior performance and cost-efficiency at scale.