Sarvam AI's 24B-parameter LLM for Indian languages Sarvam-M receives only 334 downloads in two days on Hugging Face, raising concerns about AI efforts in India
Much of Sarvam's criticism comes from comparing it to OpenAI or DeepSeek, while the problem the company is trying to solve is fundamentally different.
Analytics India MagazineMohit Pandey
Context & Ripple Effects
Sarvam emerged from stealth with funding to build LLMs supporting Indian languages, making Sarvam-M an early test of whether that localized positioning can attract an open-model developer audience. The reported download count is being read against much larger general-purpose model benchmarks, despite the company targeting a narrower use case.
The later arc shows Sarvam continuing to emphasize locally tailored models and moving toward a consumer-facing product with its Indus chat-app beta. That makes the weak initial Hugging Face reception more informative about distribution and developer discovery than a definitive measure of demand for Indian-language AI.
First-order effects
Sarvam-M’s 334 downloads in two days give critics a concrete, early signal of limited open-model uptake and intensify scrutiny of Sarvam’s India-specific model strategy.
The result pressures Sarvam to demonstrate value on the localized tasks it says it is designed for, rather than rely on headline parameter counts or comparisons with OpenAI and DeepSeek.
Second-order effects
Indian-language model builders face a higher burden to pair specialized capabilities with accessible evaluation, integrations, or product distribution; raw model release alone may not generate developer adoption.
General-purpose global models retain an advantage in attention and community momentum, so localized labs may need to reach users through applications or enterprise channels—an approach foreshadowed by Sarvam’s Indus chat-app beta.
Third-order effects
If this pattern persists, the Indian AI market may reward labs that control distribution and prove task-specific utility over those seeking validation primarily through open-model download metrics.
The episode highlights a broader split between frontier-model popularity and regional AI infrastructure: localized models can be strategically relevant, but their commercial traction will depend on deployment channels and measurable user outcomes.
The trend: Regional AI labs are shifting from model releases as standalone products toward distribution-led, application- and workflow-based proof of localized value.
re Sarvam I think with early stage companies you have no idea what they will end up becoming later, and it's actually pretty good they're getting the reps in now and understanding the lay of the land. A few years later they can have both the talent density and know how to do
In defense of https://sarvam.ai/, I will point out that there is no product we have built that was ever an instant hit. Even when we were the first mover in a new market and we had done a lot of technical work, we only got slow traction. Instant success is neither necessary
Not enough people care about Indic languages and transitively Indic LLMs. Tell me the last time we wrote and typed in our regional language. Most people who hold any kind of purchasing capability are not even comfortable reading an Indic script. I don't think there is enough
The pervasive pessimism of Deedy is a textbook case in fuckwitism. Deedy's company Glean is a competitor to Sarvam and is locking horns with Sarvam for the limited enterprise LLM udr cases. Deedy also has his own aspirations and wants to use this post to demean the work behind [i…
Great to be receiving feedback on Sarvam-M. Please keep them coming. Will help strengthen our pipelines as we start to train our sovereign model. This was particularly interesting - https://alokbishoyi.com/...
India's biggest AI startup, $1B Sarvam, just launched its flagship LLM. It's a 24B Mistral small post trained on Indic data with a mere 23 downloads 2 days after launch. In contrast, 2 Korean college trained an open-source model that did ~200k last month. Embarrassing. [image]
I like @deedydas's work but but this take misses context Sarvam-M isn't a vanity fine-tune; it's India's first open-weights 24 B Indic-centric LLM built under brutal GPU & data scarcity. Judging it by few hours of HuggingFace stats badly misses the point. Most people outside
Ok so I tested Sarvam AI's new model using some India specific tests for 1) Multilingualism 2) Cultural understanding and context 3) Empathy and decision making 4) Historica facts Here are my thoughts (Massive thread). All of these questions have thinking enabled.
Today we introduce Sarvam-M, a 24B open-weights hybrid model built on top of Mistral Small. Sarvam-M achieves a new benchmark across a range of Indian languages, math, and programming tasks, for a model of its size. Here is a detailed technical blog on how we customize [image]
This response definitely makes me want less to try it now. People can take it seriously if you explained how it solves a pain for them instead of talking in platitudes. Make it relatable. It's hard to understand what Sarvam has achieved right now, beyond national pride.
We all want to root for Sarvam. But wanting isn't enough when you're selling duty and technical jargon instead of desire. It's a narrative problem. Right now, my perception of Sarvam's unsaid positioning feels like “We've done some cool technical stuff. Support us because we're
Good point, IMO People find Sarvam unreliable because users have to translate what Sarvam can do in their mind I am sure Sarvam with hundreds of $ mn raised has super smart growth hackers but my $0.02 Starting from an ICP should answer all questions on Sarvam's “right to win”