Tech Explained: Here’s a simplified explanation of the latest technology update around Tech Explained: Has Sarvam AI beaten ChatGPT and Google Gemini? Yes and no! in Simple Termsand what it means for users..
It’s not every day that an Indian AI startup grabs global attention. So when Sarvam AI made headlines claiming it had beaten ChatGPT and Google Gemini, the excitement in India was understandable. Articles like “India’s Sarvam AI beats Google Gemini and ChatGPT, the world is impressed” quickly went viral.
But once the initial buzz settled, a more important question emerged: has Sarvam AI truly outperformed the world’s best AI models—or is the story more nuanced?
The answer is both yes and no.
On February 5, Sarvam AI cofounder Pratyush Kumar announced that Sarvam Vision topped the olmOCR-Bench, a benchmark that measures optical character recognition (OCR). OCR tests how well AI models can read scanned documents, handwritten text, complex fonts, tables, and mathematical notations.
Sarvam Vision scored 84.3% accuracy, beating OpenAI’s ChatGPT, Google’s Gemini 3 Pro, and even China’s DeepSeek OCR v2. On OmniDocBench v1.5, it scored an impressive 93.28%, excelling at complex layouts, technical tables, and formulas.
Sarvam Vision is trained specifically on Indian scripts and writing styles. That makes it far better at handling regional languages, mixed-language documents, and Indian formatting quirks—areas where global models are capable, but not deeply fine-tuned.
This makes Sarvam Vision extremely useful for Indian businesses dealing with scanned forms, government documents and multilingual paperwork—often at a lower cost than foreign AI models.
Sarvam’s second big win is Bulbul V3, its text-to-speech model. Bulbul outperforms global leaders like ElevenLabs when it comes to Indian accents, pronunciations and speech patterns. Again, the advantage comes from focused training on Indian linguistic realities.
Here’s where the “no” comes in.
Sarvam’s models are not general-purpose AI systems. They don’t compete with ChatGPT or Gemini in everyday AI use—things like reasoning, coding, tutoring, creative writing, or multimodal problem-solving.
ChatGPT can help you solve a JEE-level physics problem step-by-step. Gemini can generate mock exams and guide students interactively. These are tasks Sarvam’s models aren’t designed for.
In short:
• ChatGPT and Gemini are jacks of all trades
• Sarvam AI is a master of a few, highly specific ones
That difference is reflected in scale. Sarvam Vision has around 3 billion parameters, while Google Gemini 3 is rumoured to have nearly 2 trillion. Bigger models generally mean broader intelligence—but also require massive compute, data centres, and hundreds of thousands of GPUs.
Even with these limitations, Sarvam AI’s achievement is significant.
Vision and Bulbul are proof of capability. They show that Indian startups can build world-class AI tools from scratch when they focus on well-defined problems. The constraint isn’t talent—it’s access to compute and infrastructure.
By beating global models in specific benchmarks, Sarvam AI has made an important point: with the right focus, Indian companies can lead, not just follow.
That alone makes Sarvam Vision and Bulbul worth celebrating!