Google just promised up to 200 European researchers direct access to five of DeepMind’s most serious science models, plus $3 million in cloud credits to run them. The headline sounds huge. The details are thinner. Here is what each tool does, what Google has not said yet, and how you can try most of them this week from anywhere in the world.
On October 14, 2026, Maureen Costello, a Google Cloud vice president, announced the pledge in a post on Google’s blog. I read it next to each tool’s own documentation, because a list of Alpha names tells you nothing until you know what each one does and who is allowed to touch it.
What Google Actually Announced
Google DeepMind and Google Cloud will give up to 200 top European scientific researchers direct access to their frontier models, together with $3 million in compute credits. The named tools are AlphaFold 3, WeatherNext, AlphaGenome, AlphaEvolve and AlphaEarth Foundations. Access runs through RAISE, the European Commission’s AI for science initiative.
Quick math, and this one is mine, not Google’s: split evenly, $3 million across 200 researchers is about $15,000 each. Google has not said the credits will be divided that way.
| Tool | What it does | Where you can already try it |
|---|---|---|
| AlphaFold 3 | Predicts the structure of proteins, DNA, RNA, ligands and ions, and how they interact | AlphaFold Server, free for non-commercial research |
| AlphaGenome | Predicts how DNA variants change gene regulation, from sequences up to 1 million letters | AlphaGenome API, free for non-commercial use |
| WeatherNext | AI weather forecasting that produces many possible outcomes from one starting point | Earth Engine, BigQuery, Vertex AI, plus the Weather Lab site for experimental models |
| AlphaEvolve | A Gemini-powered agent that evolves code to find better algorithms | Google Cloud, generally available since July 2026 |
| AlphaEarth Foundations | Compresses satellite data into 10 meter map embeddings | Satellite Embedding dataset in Earth Engine |
The Five Tools in Plain English
AlphaFold 3 is the famous one. Google says it improves prediction of protein interactions with other molecule types by at least 50 percent compared with existing methods. If you work in biology or drug research, this is the tool you already know.
AlphaGenome reads long stretches of DNA and predicts what a mutation does to gene regulation. According to DeepMind’s announcement, it beat the best external models on 22 of 24 single sequence evaluations and matched or beat them on 24 of 26 variant effect evaluations. The work was published in Nature in January 2026.
WeatherNext is a family of forecasting models. WeatherNext 2 is eight times faster than its predecessor and delivers forecasts at up to one hour resolution. Google says it can generate hundreds of possible weather outcomes from a single starting point, including rare and severe ones.
AlphaEvolve is the odd one out, because it is not about biology or weather. It is a coding agent that pairs Gemini models with automated evaluators and an evolutionary loop. Inside Google it sped up a key kernel in Gemini’s architecture by 23 percent, which cut Gemini’s training time by about 1 percent.
AlphaEarth Foundations turns a flood of satellite data into compact map layers. Each 10 meter pixel gets a 64 dimension summary of one year of surface conditions. In DeepMind’s paper, it had a 23.9 percent lower error rate on average than competing approaches, as reported by IEEE Spectrum.
Why This Matters Even If You Are Not a Researcher
Here is the thing. Most of these tools are not secret. AlphaFold Server, the AlphaGenome API and the AlphaEarth dataset are already public for research use. What the pledge adds is compute credits, a direct line to the models, and a government channel that gives it credibility.
My read is that the scarce resource for most researchers is rarely the model itself. It is the cloud bill and the weeks lost working out how to get access. Credits attack the second problem, and that is why companies hand them out. A lab that builds its pipeline on one platform tends to stay on it.
The scale tells its own story too. The EU launched the RAISE pilot with €107 million in Horizon Europe funding, and Horizon Europe is also putting €600 million toward dedicated access to AI gigafactories. Next to that, Google’s $3 million is small. The value is in the models, not the money.
Try AlphaEarth Today: A 10 Minute Walkthrough
You do not need to be in Europe, and you do not need to wait for Google. The AlphaEarth Satellite Embedding dataset sits in Google Earth Engine, which is free for research, education and nonprofit use. Let’s break it down.
- Register for Earth Engine. Use the sign up link on the dataset page. Free access covers research, education and nonprofit work.
- Open the dataset page. Search the Earth Engine catalog for Satellite Embedding V1, then click the button that opens the sample script in the Code Editor.
- Run the sample. It loads annual embedding layers, picks a point, and compares 2023 against 2024 using a dot product.
- Swap in your own location. Find the line that creates the point and replace the numbers with your own longitude, then latitude. Longitude comes first, and mixing them up is the most common mistake.
- Read the map. The script labels the layer so that brighter pixels mean less similarity between the two years. Bright patches are places that changed.
- Check what you found. Switch to the satellite view and zoom in on a bright patch. The embeddings tell you where something changed. They do not tell you why.
Try AlphaFold 3 in Four Steps
The EMBL-EBI training guide walks through AlphaFold Server, and it is the clearest explanation I found.
- Sign in to AlphaFold Server with a Google account.
- Add your molecules. For a protein, paste the single letter amino acid sequence. You can add DNA, RNA, ions and other supported molecules to model them together.
- Confirm and submit the job. You get a limited number of jobs per day, and you can save extra jobs and run them after your quota renews.
- Open the result and check the confidence scores before you trust any structure.
For the other three: AlphaGenome has a free API for non-commercial use, Weather Lab lets anyone browse experimental AI weather models, and AlphaEvolve is a Google Cloud service that needs a seed program and a way to score results.
Honest Limitations
- The rules are missing. The Next Web reports that Google has not said how researchers will be chosen, when access starts, or how the credits will be split.
- $3 million goes only so far. Spread across 200 researchers, it averages about $15,000 each, which points to experiments rather than large scale training. Analytics Insight notes that teams may need grants or other funds once the credits run out.
- Most access is for non-commercial work. AlphaFold 3 weights are limited to non-commercial organizations. The AlphaGenome API is non-commercial too, and DeepMind says its predictions are for research use, not clinical decisions.
- The models have known gaps. DeepMind says AlphaGenome still struggles with regulatory elements more than 100,000 letters away and was not designed for personal genome prediction. WeatherNext’s experimental models in Weather Lab are not official forecasts.
- AlphaFold’s team has changed. The Decoder reports, citing the Financial Times, that DeepMind broke up the team behind AlphaFold and reassigned most of the original paper’s authors. DeepMind confirmed the moves. The tool is still on Google’s list, so treat this as context for long term planning, not as a warning.
- It is Europe only. If you work elsewhere, the pledge does not apply to you. The public versions above still do.
Your Move
If you are a European researcher, bookmark Google’s announcement and the Commission’s RAISE page, and write a one page project brief now. When applications open, the lab with a clear question and a data plan will move first.
If you are not, pick one tool from the table and spend an hour with it this week. The Earth Engine walkthrough is the fastest start. More coverage like this lives in my Google and Research categories.
Know a researcher, a grad student or a data nerd who would want this? Send them this post, and tell me in the comments which of the five tools you would try first.
Sources
- Google: We’re powering European scientific discovery with advanced AI compute and frontier models (October 14, 2026)
- The Next Web: Google DeepMind opens its science AI models to 200 European researchers
- eeNews Europe: EU Commission launches RAISE virtual AI institute with €107m
- European Commission: Resource for AI Science in Europe (RAISE)
- Google DeepMind: AlphaFold 3 predicts the structure and interactions of all of life’s molecules
- The Decoder: Google DeepMind releases AlphaFold 3 for scientific use
- EMBL-EBI: AlphaFold Server, your gateway to AlphaFold 3
- The Decoder: DeepMind dismantles its AlphaFold team
- Google DeepMind: AlphaGenome, AI for better understanding the genome
- Nature: AlphaGenome paper
- GitHub: AlphaGenome API
- Google DeepMind: WeatherNext 2
- 9to5Google: WeatherNext 2 announcement coverage
- Google DeepMind: Weather Lab
- Google Cloud: AlphaEvolve on Google Cloud
- Google Cloud: AlphaEvolve is available for everyone
- Google Earth Engine: Satellite Embedding V1 dataset
- IEEE Spectrum: Google DeepMind AlphaEarth Foundations
- Analytics Insight: Google gives 200 European researchers access to advanced AI models
Leave a Reply