Google DeepMind and Google Research today announced WeatherNext 2, the new generation AI prediction system, which promises to deliver faster, more accurate, higher-resolution probabilistic weather forecasts. It is also being integrated into Google products and APIs for developers.
Based on decades of machine learning research in ensemble forecasting, WeatherNext 2 is designed to rapidly and accurately offer a wide range of future scenarios, which is what matters most for predicting the probabilities of low but significant events like tropical storms, intense downpours, or even extreme winds.
What is WeatherNext 2?
WeatherNext 2 uses an AI-powered forecasting system that generates weather predictions up to 8 times faster than previous models and hundreds of potential future scenarios. This approach helps better understand uncertainty and dramatically improves forecasts of rain, wind, and other changing weather conditions.
Google has already begun to integrate WeatherNext 2 with:
- Google Search
- Gemini
- Pixel Weather
- Google Maps & Maps Platform’s Weather API
Users and developers will soon be receiving newer, higher-resolution forecasts with more certainty throughout Google products.
What makes WeatherNext two different?
There are three significant developments in WeatherNext 2.
- Scale and Speed: Google says WeatherNext 2 can produce forecasts about 8x faster than predecessors, allowing more scenarios to be simulated per forecast cycle. Inference speed is faster, which means more ensemble members are generated in the same period, and the addition of more members of an ensemble generally results in higher estimations of uncertainty for the infrequent events.
- High-Resolution, Probabilistic Outputs: WeatherNext 2 produces forecasts with up to an hour resolution and can provide hundreds of potential outcomes, giving Forecasters and downstream service providers a broader perspective on possible future outcomes rather than just a single, deterministic forecast—the importance of probabilistic framing to decision-making in fields such as transportation, energy, and emergency management.
- Integrations for Operational use: Google is already offering WeatherNext 2 outputs into consumer and developer-facing surfaces: Search, the Gemini assistant, Pixel Weather, and the Google Maps Platform Weather API. The company has announced that integration into Google Maps (public-facing map weather layers) is expected in the next few weeks. This isn’t an experiment for the lab, but the model is currently being utilised throughout Google’s ecosystem.
Why WeatherNext 2 Is Good?
1. It can generate forecasts 8x faster.
Traditional weather models rely on physics-based models that could take several hours.
WeatherNext 2 utilises AI to create forecasts in a fraction of the time.
What makes this good:
- You’ll get more recent, up-to-date weather forecasts.
- Important short-term events such as storms, rainfall surges, or wind changes can be anticipated earlier.
- Faster forecasts mean frequent updates across the Google ecosystem.
2. It can provide hundreds of possible future results
Most apps offer only one forecast, but it’s not always the case.
WeatherNext 2 creates many scenarios in the same time frame.
What makes this good:
- Provides a more accurate understanding of the uncertainty.
- Enhances the detection of rare but significant events such as flash flooding or cyclones.
- Aids emergency teams and businesses make better choices.
3. It increases the accuracy of forecasts for short-term times.
Short-term weather (the next several hours) is the most difficult to predict, particularly the timing of rain.
AI is exceptionally adept at interpreting local atmospheric patterns and spotting signals before they become apparent.
What makes this good:
- More precise times for rain start/stop.
- Improved hourly-by-hour time-to-hour data for travel, maps, deliveries, as well as outdoor tasks.
- There is less chance of unexpected weather blips.
4. It decreases the computational cost.
WeatherNext 2 can run efficiently using the latest AI accelerators rather than supercomputers.
What makes this good:
- Costlier to run.
- Provides regular forecast updates.
- It makes large-scale ensemble forecasting available.
5. The feature is integrated now into the real Google products.
This isn’t an experiment.
WeatherNext 2 has already started generating power:
- Google Search weather
- Gemini weather-related responses
- Pixel Weather
- Google Maps Platform Weather API
- Soon to be available: Google Maps weather layers
The reason this is beneficial:
Millions of people will benefit from this without having to download a special software or application.
6. It is an exciting future for hybrid forecasting for weather.
WeatherNext 2 is compatible with traditional models of physics, not against them.
What makes this good:
- Combines the speed of AI and the reliability of well-established scientific models.
- This opens the way to more effective early warning systems worldwide.
- Improves forecasting in areas that have limited weather data.
Research Lineage from GenCast to WeatherNext
Image Source – Google
WeatherNext 2 builds on ongoing research by DeepMind and Google Research into ML-based ensemble forecasting (GenCast and similar models). Previous work showed that learned, generative methods can surpass traditional numerical weather prediction (NWP) ensembles across numerous verifiability metrics and can be executed more efficiently with modern accelerators.
DeepMind’s previous GenCast work and later tests have shown that machine-learning ensemble forecasts can be very competitive with, and even better than, physics-based systems.
Why Probabilistic, Many-Scenario Forecasts Matter
Traditional deterministic forecasts offer a single “best estimate,” which can be helpful but is incorrect when the weather is turbulent. Ensemble forecasting, which is the process of running multiple simulations with slightly different starting conditions or model variations, is the most common method for assessing forecast uncertainty.
WeatherNext 2’s ability to generate hundreds of possible futures increases the ensemble advantage, allowing forecasters not only to see the most likely scenario but also to consider the likelihood of extreme outcomes. It can alter decisions on evacuations, power trading, supply chain rerouting, and even infrastructure operations.
Use cases from the real world and industry consequences.
- Forecasting for Consumers: Pixel Weather and Google Search weather cards, and Gemini are now using surface forecasts powered by WeatherNext technology, which means millions of users will be able to access forecasts generated by the model in their daily situations.
- Businesses and Developers: The Google Maps Platform Weather API already provides AI-based forecasts. WeatherNext 2 will enhance the quality of information available to developers developing logistics plans for retail and tools to manage energy. Google previously focused on enterprise use cases for energy traders and logistics companies when it launched WeatherNext.
- Preparedness for Disasters: DeepMind has run experiments using cyclone forecasts and worked with meteorological organisations to test scenarios. Faster ensemble forecasts that better identify high-impact tracks with low probability or changes in intensity could lead to earlier warnings and a better allocation of resources. However, translating research results into operational warnings requires collaboration with national meteorological agencies and careful verification.
Strengths and the Remaining Limitations
WeatherNext 2 is a significant step forward; however, it’s not a replacement for physics-based NWP systems. The most important thing to remember:
- A complement, not a Substitute: ML-based ensembles excel in speed and specific accuracy metrics; however, physics-based NWP remains vital for long-term forecasts and for providing physically consistent, conservation-aware solutions in certain situations. Google puts WeatherNext 2 as a powerful supplement that can improve probabilistic ability and speed.
- Operationalisation and Validation: DeepMind has published research and worked with organisations to evaluate its methods; however, widespread adoption requires independent verification across seasons, regions, and rare events. The data gap, the regional biases and edge cases need careful testing before they can be used in vital warning systems.
- Reproducibility and Transparency: As with all ML forecasting systems, transparency regarding training data, evaluation protocols, and limitations is crucial for confidence. DeepMind’s research pages on the internet and blog posts provide a summary of the results; however, independent peer review and public benchmarks are essential next steps.
The Future of WeatherNext 2
WeatherNext 2 is not just an update; it is the basis for a larger change in how global weather forecasting is evolving. In the coming years, the technology is likely to impact the consumer experience, scientific forecasting systems, and global climate-related decision-making.
Below is a thorough, forward-looking overview of what WeatherNext 2 will likely entail.
1. Integration with Real-Time Global Weather Systems
WeatherNext 2 continues to evolve to become a multi-faceted forecasting tool that is integrated with:
- National meteorological authorities
- Satellite data providers
- Global observation systems for global observation
As Google improves its AI capabilities for ensemble forecasting, WeatherNext 2 could become an integral part of operational forecasting pipelines, helping both national and local weather agencies identify storms more quickly and conduct risk assessments more confidently.
2. Better Prediction of Extreme Weather Events
AI ensemble models such as WeatherNext 2 are perfect for:
- Short-term rainfall bursts
- Cyclone intensity shifts
- Sudden wind shifts
- Flash conditions of flooding
- Rapid storm formation
The future versions are likely to:
- Take care of extremes with greater precision
- Find out about rare events earlier
- Give more specific impact pathways
This can transform the way disaster teams respond by allowing emergency personnel to respond ahead of time when they did before.
3. Ultra-Localised, Street-Level Forecasting
As the computing power gets more efficient, WeatherNext 2 is capable of:
- Forecast weather with much finer resolutions
- Offer neighbourhood-level insights
- Improve the delivery of hyperlocal navigation and travel plans
Google Maps, ridesharing services, and logistics platforms can make real-time routing adjustments based on weather changes.
4. Smarter Probabilistic Forecasting
The main strength in WeatherNext 2 lies in its ability to model hundreds of future scenarios.
Future versions could:
- The number of simulations by thousands
- Combine atmospheric physics and AI learning in real-time
- Offer confidence scores for every incident (e.g. the probability of rain spike analysis)
This will allow companies and government agencies to use weather data more systematically.
5. Climate-Scale Forecasting
Although WeatherNext 2 is primarily focused on weather for the short-term, Future systems could evolve into:
- Models for seasonal prediction
- Climate anomaly predictors for climate anomalies
- Tools for long-term risk assessment
This can help sectors such as agriculture, energy, insurance and coastal management.
6. Smarter Consumer Experiences Across Google Products
As WeatherNext 2 continues to evolve, Google may introduce:
- Scores of reliability for precipitation hour-by-hour
- Smarter notifications for Pixel Weather
- The weather forecasts are based on scenarios and can be found in Gemini
- Weather and travel advisories on Google Maps
- Smart home automation based on weather devices
Forecasting for weather will change from “what will happen” to “what scenarios are likely and how you should respond.”
7. Industry Adoption via the Google Maps Weather API
Enterprises and developers using the Google Maps Platform will gain:
- Higher-resolution data
- Weather-based forecasting for routing
- Risk-aware delivery planning
- Forecasting on-demand, probabilistic and on-demand
Future updates could include:
- Plug-and-play weather risk module
- Automated hazard detection
- Weather intelligence specific to industry (energy, shipping, aviation)
What does this mean for the Developers and users?
If you’re a developer who uses the Google Maps Platform Weather API, expect higher-resolution hourly estimates and more accurate uncertainty estimates in the near future. Customers who use Pixel Weather and Google Search will see forecasts generated by WeatherNext technology, which will enhance the short-term timing and intensity of wind and rain forecasts.
For logistics and energy trader teams, the speed increase and richness of the ensemble can lead to more precise risk management and automated decision-making.
Final Thoughts
WeatherNext 2 is a clear improvement on DeepMind, along with Google’s efforts to enable machine-learning ensembles on an operational scale. With its faster inference speed, more robust probabilistic outputs, and integration with Google’s developer and consumer services, WeatherNext 2 accelerates the trend towards AI-augmented forecasting.
The full impact of the model will depend on the ongoing validation process, clear assessment, and close cooperation with meteorological agencies. However, the model’s speed and ability to work in groups provide a notable advance for both daily weather forecasts and high-risk decisions in the event of extreme weather.
FAQs
1. What date was WeatherNext 2 released?
Google DeepMind and Google Research announced WeatherNext 2 on November 17, 2025.
2. How fast is WeatherNext 2?
Google reports that WeatherNext 2 is about 8x faster than previous versions, enabling many more scenario simulations per forecast.
3. What is WeatherNext 2 currently being utilised for?
Model outputs power weather forecasts in Google Search and Google Maps, the Gemini assistant Pixel Weather, and the Google Maps Platform Weather API. Google says weather layers in Google Maps are scheduled to be upgraded over the next few weeks.
4. Developers can directly have access to WeatherNext 2?
Developers can access AI-powered weather data via the Google Maps Platform Weather API. Google Maps Platform Weather API. Google’s developer blog and documentation provide evidence that WeatherNext technology is being integrated into the API; however, direct access to model data (weights and hosted endpoints) is not typically distributed in the same way as a publicly released model. Go through the Maps Platform docs and Google announcements for the most up-to-date access information.
5. Can WeatherNext 2 replace the national forecasts or models that are traditional?
No. Google frames WeatherNext 2 as a tool to enhance forecasting skills and speed, as well as to improve the existing NWP system. Operational warning systems remain predominantly controlled by national meteorological agencies, which could incorporate ML forecasts alongside physics-based models following validation.
