What Happened

DeepMind, Google’s AI research unit, has officially launched WeatherNext 3, the third iteration of its AI-powered weather prediction model. The announcement came via a technical blog post and a developer keynote at the Google Climate AI Summit. Unlike earlier versions, WeatherNext 3 integrates a hybrid architecture combining transformer neural networks with diffusion-based probabilistic generators, enabling it to output not just a single forecast but a range of possible outcomes.

The model is trained on roughly 60 years of ECMWF reanalysis data and incorporates real-time satellite feeds. In internal benchmarks, it outperformed the previous WeatherNext generation by 35% in temperature prediction and 42% in precipitation event detection. For India, the model refines forecasts for the monsoon season, which is critical for over 60% of the country’s farmland. It also improves early warnings for cyclones along the Bay of Bengal and Arabian Sea, reducing false alarms by 28% in test cases.

Why It Matters for Businesses and Developers

For Indian businesses, the ability to access hyperlocal weather intelligence is a game-changer. Insurance companies in Chennai and Mumbai can use probabilistic forecasts to price monsoon-linked crop policies with greater precision. Renewable energy operators, especially those managing solar parks in Tamil Nadu and Rajasthan, can now predict cloud cover and solar irradiance more accurately, optimizing grid load balancing. Supply chain startups managing cold chains for agri-products can preempt weather disruptions and reroute logistics, saving crores per event.

Developers and data science teams can integrate WeatherNext 3 via a simple API, eliminating the need to maintain complex weather models. The model offers high-resolution output without a supercomputer, reducing compute costs by an estimated 90% compared to running operational NWP models. Early access partners in India, including a private weather services firm in Hyderabad, reported a 50% faster turnaround for client forecasts. This opens doors for small startups to build weather-aware applications in agriculture, logistics, and climate-risk analytics, which previously required heavy infrastructure investments.

Availability and Rollout

WeatherNext 3 is now in public beta for developers globally, including India. It is accessible through the Google Cloud Vertex AI platform and the Google Maps Platform Weather layer. The API provides both historical and real-time forecasts, with scheduled updates every 15 minutes for major metros like Mumbai, Delhi, Bangalore, and Chennai. Pricing starts at $20 per 1,000 API calls, with a free tier of 5,000 calls per month for developers. No special eligibility is required; sign up via the Google Cloud Console. A general availability release is slated for Q4 2025, with on-prem deployment options for government agencies.