Today we are launching Dhenu Vision: the vision intelligence API for agriculture. State of the art for identification of pests and diseases in real farm photos. 95.1% precision, where the best frontier model scores 84.6%.

Up to 40% of the world's harvest is lost to pests and disease. Half a billion farming families carry that risk. Most of it starts with one missed or wrong diagnosis.

A real plant rarely has one problem. Traditional AI identifies one issue at a time. Frontier vision language models hallucinate confidently. Dhenu Vision finds every issue: pest, disease and weed together, each boxed, each confirmed by a specialist model.

It returns what a decision needs: - the finding, located on your image - calibrated confidence - advice you can act on - an answer to your own question about the photo

Coverage on day one: 85 crops, 986 pest, disease and weed issues, plus livestock and aquaculture. Every result localized end to end in 63 languages. Hindi to Swahili, names through advice.

Built in India, benchmarked against the world. The debate is whether India can build frontier AI. In the layer that feeds people, we are not catching up. We are leading. Start today:

We end up creating an agricultural diagnosis benchmark from scratch. Will be releasing the benchmark, results, and whitepaper, next week.

Don’t you think a phone app will help small farmers in India? Of course with affordable monthly subscription.

We are working on a free one. Releasing soon.

Problem was never the model. It is who will capture the image. Will it be human, drone or jigbased overhead cams.

Please release the benchmarks to public and if possible, publish scores of Fable & GPT 5.6 Sol as well.

Yes, Next week.

Domain specific vision is where generic models start losing their edge.

Congratulations @chheplo , this is quite Interesting! I published a book chapter on Precision Agriculture, I want to understand this better, can we connect? link -

Looks awesome

Brilliant!

@fpingham 👀

The product layer now matters: show confidence, ask for a second image and route doubtful cases to an expert. That is how a strong model becomes a tool people can trust.

Its a good model sir

Congratulations 👏🏻🙏🏻 Thanks for reminding us to care for the roots and bringing technology to the farmers.

If only you can bring collective farming, India will be one of the most self-sufficient economies along the lines of China in food grains. We need one farming policy, one crop system and a clear cultivation policy with markets, then rewarding purchases with price points and open exports so they get good returns. Marginal farming has been the biggest curse on the economy. AI should be able to collectively decide a uniform cultivation strategy.

@vijayshekhar it'd be amazing if I could use PayTM as superapp in my farm to check this. Apart from getting credit.

Man-o-man! This is huge

Much needed. Should also interlink into the data of efficacy proven solutions so the MRLs are below which help not just the farmer but also the consumer. All the best @chheplo

Awesome work!

A 10.5% gap in real-world farm photos. This is exactly why specialized AI beats generalists every time. Frontier models chase broad benchmarks but they lose in the trenches where specificity matters.

how can we get vision API access, could not find any info in your website for API.

and

What is happening with my timeline!! Kudos man! May Prabhu Shree Ram guide you to the path of least resistance

amazing model specifically for diseases and pest superb excited about the implementation

Distribution is the key... To reach every farmer .. isn't ?

This looks awesome! And thank you for not using an over the top BGM.

@um_aadi

seeds and soil. both corrupted. soil can be there too. lots of cheap tools to monitor soil health.

The j curve in indian R and D has begun

Congratulations 👏🎉👏👏 💐💐💐

Man I am feeling too jealous .. Since a month or so daily I am seeing 2-3 new startups doing this or that... Are we going to grow exponenyin next decade ?? WTF is going on ???

We need to think about how to run this on edge devices without internet

Wonderful work, Pratik! Check this out, @hareshsanghavi.

Benchmarks?

There was none so we created one and will be publishing the benchmark and complete results next week

just curious is there any free access for testers?

Yes

Kudos Pratik. Do you have idea about any vision model that can check a heap of Paddy or Wheat in an APMC, and then tell about moisture, black seed, stone mix etc. to Determine quality and can help with price. Such a model could do wonders. My father need one.

🎮

Hey, any info about the dataset? Is it os

Keep up the good work