The Indian government just launched three new artificial intelligence models through its IndiaAI Mission. The main model, EKA-1, contains 26.3 billion parameters and was trained on 7.5 trilli
The Indian government just launched three new artificial intelligence models through its IndiaAI Mission. The main model, EKA-1, contains 26.3 billion parameters and was trained on 7.5 trillion tokens. It covers 22 Indian languages, along with English. The government also released two smaller versions, Dhīr 2B and Dhīr 5B. These two models are fully open source and support 10 Indian languages.
▶️IndiaAI Mission releases Indigenous Foundation Models EKA-1 by Soket and Dhīr 2B & Dhīr 5B by ZenteiQ Under the IndiaAI Innovation Centre Pillar
▶️EKA-1 by Soket Features 26.3 Billion Parameters Trained on Approximately 7.5 Trillion Tokens Across All 22 Scheduled Indian Languages and English for Math, Code, Reasoning & Agentic Use
▶️ZenteiQ Introduces Open-Source Dhīr 2B and Dhīr 5B Multilingual Assistants Supporting English and 10 Indian Languages for Secure, On-Premise, Offline-Capable Deployment Across Critical Sectors
Read the complete update here: https://t.co/57Rytr6GIC
— PIB India (@PIB_India) November 4, 2026
Developers specifically built the Dhīr models to run entirely offline on standard mobile phones. Because the software operates directly on the physical device, it processes data locally and keeps all user information completely off the internet.
Technology users quickly analyzed the practical benefits of the release. On X, @Zangokoon, a Quantitative trader, added that having a small, locally run model on a phone is appealing, although they questioned its performance and confirmed its open-source status.
@ahsuzm 手机本地小模型，还是挺有吸引力的，不知道表现怎么样，是否开源
— Gokoon (@Zangokoon) November 4, 2026
A Chinese automotive technology engineer on X, @Charge, explained that running a 5B model on a phone with 6GB of memory requires 4-bit quantization. They compared this to modern car systems that use small offline AI models to handle basic commands when internet access is unavailable.
印度也憋了个大招！
11月3日，IndiaAI Mission放了三个国产基础模型：EKA-1有263亿参数，约7.5万亿token喂出来，22种印度语言加英语全拿下；Dhīr 2B和5B开源，英语加10种印度语言，离线能在手机上跑，数据不出设备。
#印度AI #大模型 #开源 https://t.co/3K98m7q5pi
— 芝诺Zeno（AI+薄肌版） (@ahsuzm) November 4, 2026
@An_yhl agreed with the model, describing the ability to run programs offline on a phone as a highly practical feature.
@ahsuzm 离线能在手机上跑，这点还挺实用的。
— 晚晚 (@An_yhl) November 4, 2026
READ ALSO: India’s IT Giants Face Another Weak Quarter as AI Pushes Prices Down
Standard artificial intelligence apps pose a security threat to cryptocurrency traders. If an investor asks the AI to review a private financial plan, check a smart contract, or analyze a crypto wallet address, that sensitive data travels across public internet networks. Hackers constantly monitor these networks to steal financial information while it is in transit.
The Dhīr 5B model removes this internet risk entirely. It uses 4-bit quantization to compress massive amounts of computer code so it fits easily inside a standard phone’s limited memory chip. Because the model fits on the phone, the device does all the mathematical processing locally. A crypto investor can download a decentralized application that uses this offline AI to scan crypto trades for hidden scams.
The phone reads the data, analyzes the risk, and gives an answer without ever connecting to a Wi-Fi or cellular network. This guarantees that private trading strategies and wallet details never leave the user’s physical hands.
The IndiaAI Mission released the Dhīr models as open-source software. The government published the raw, foundational code online for anyone to read, copy, and test. This exact level of public transparency is the core foundation of the entire cryptocurrency industry.
In traditional technology, large corporations hide their artificial intelligence inside private black boxes. By making the AI open-source, India treats artificial intelligence the exact same way developers treat Bitcoin. Blockchain engineers can review the Dhīr code line by line.
They can mathematically verify that the software contains no hidden spyware or government backdoors, and that they can safely build it into decentralized trading apps. This public verification process ensures that the AI serves the user, rather than secretly working for a central tech corporation.
Meanwhile, Microsoft Chairman and CEO Satya Nadellasaid India could have the world’s largest developer community by 2030, showing the country’s growing pool of technology workers as a major advantage in the AI race.
The global cryptocurrency market operates almost entirely in English. Most digital wallets, trading platforms, and decentralized finance applications do not support local dialects. This language barrier blocks billions of people in rural areas from accessing alternative financial tools. They cannot buy Bitcoin or trade digital dollars simply because they cannot read the on-screen menus.
The EKA-1 model directly addresses this exclusion by natively processing 22 Indian languages. Unlike basic software that only translates English words, this model understands the grammar and cultural context of regional languages. Cryptocurrency developers can integrate this native language model directly into their trading platforms.
A local farmer or small business owner can then use voice commands in their own language to interact with a blockchain network. They can ask the app how to secure a crypto loan or how to send digital money across borders. By removing the English requirement, this technology allows millions of new users to finally enter the global digital economy.
Enjoyed this? BookmarkDeFi Planet, explore related topics, and follow us onTwitter,LinkedIn,Facebook,Instagram,Threads, and CoinMarketCap Community for seamless access to high-quality industry insights.
Take control of your crypto portfolio with DEFI PLANET PRO, DeFi Planet’s suite of analytics tools.
The post India Launches Three New AI Models Supporting Local Languages appeared first on DeFi Planet.
