AlphaGo Zero

Closed weights DeepMind 46.4M parameters October 2017

No estimate

No hardware requirements for this model

The weights for this model have not been published, so it cannot be downloaded or run on your own hardware at any size. It is reachable only through its provider, and no graphics card changes that.

On record

Full specification

Everything on record for this model. Most of it describes how it was trained rather than how it runs — useful context for judging how much work went into it, and how it compares with models built at a different scale.

Origin

Who built this model, where, and when it was published.

Organisation
DeepMind
Organisation type
Industry
Country
United Kingdom of Great Britain and Northern Ireland
Published
18 October 2017
Authors
D Silver, J Schrittwieser, K Simonyan, I Antonoglou

What it does

The problem areas the model was built for. A model can carry several of each.

Domain
Games
Task
Go
Approach
Self-supervised learning

Size

How large the model is and how much data it was trained on. Parameters are the figure that decides whether it fits on a given graphics card.

Parameters
46.4M

Input size: 19*19*17=6137, internal dimension 19*19 1 conv block 17*3*3*256=39168 39 residual blocks 2*3*3*256*256=1179648 Policy head 256*2+2*19*19*(19*19+2)=262598 Value head 256*1+1*19*19*256+256*1=92928 Total: 39168+39*1179648+262598+92928=46400966

Training data
6,348,800,000 tokens

"Over the course of training, 29 million games of self-play were generated" Approx 200 moves per Go game on average https://homepages.cwi.nl/~aeb/go/misc/gostat.html Thus 200 * 29e6 = 5.8e9

Training compute

The arithmetic performed to train the model, measured in floating-point operations. It is a measure of what the training run cost, not of how fast the finished model answers you.

Training compute
6.5 × 10²⁰ FLOP

Updating this compute estimate to only account for direct training compute, not synthetic data generation compute. Number of connections (to calculate forward FLOP) Input size: 19*19*17, Assuming internal dimension stays at 19*19 1 conv block 19*19*17*3*3*256=14139648 40 residual blocks 19*19*2*3*3*256*256=425852928 Policy head 19*19*256*2+2*19*19*(19*19+2)=446918 Value head 19*19*256*1+1*19*19*256+256*1=185088 Total: 14139648+40*425852928+446918+185088=17048888774 Forward FLOP: 2*17048888774=3…

How it was established
Third-party estimation,Hardware

The training run

What it physically took to train: which chips, how many, for how long, and what that drew from the wall.

Training hardware
Google TPU v1
Wall-clock time
480 hours (20 days)
Compute cost
$1,214

Availability

Whether you can obtain the model and run it on your own hardware, which is what decides if any of the graphics-card figures on this page apply.

Weights
Closed — provider access only
Model access
Unreleased
Training code
Unreleased

How it is classified

Labels the source dataset applies when tracking notable models, and how confident it is in the entry.

Frontier model
Yes
Why it is tracked
Highly cited
Record confidence
Confident
Citations
10,021

Sources

Where this record came from and when it was last checked.

Reference
Mastering the game of Go without human knowledge
Last updated
1 January 2026

What the numbers mean

Where it came from

AlphaGo Zero was published by DeepMind, in United Kingdom of Great Britain and Northern Ireland, in October 2017. It comes out of industry.

It works in Games, and is recorded as doing go.

This is a closed model: the trained values stayed with whoever produced them, and there is no local version to run.

How it was trained

The training run consumed about 6.5 × 10²⁰ FLOP, on Google TPU v1. That figure describes the cost of creating it and has no bearing on how quickly it generates text.

The training set ran to roughly 6,348,800,000 tokens.

Its inclusion criterion is highly cited.

Answers

AlphaGo Zero — common questions

01

How much compute was used to train AlphaGo Zero?

Around 6.5 × 10²⁰ FLOP, on Google TPU v1. That measures what producing the model cost and says nothing about how quickly it answers once trained — inference speed comes from memory bandwidth, not from the training budget.

02

What GPU do I need to run AlphaGo Zero?

None. AlphaGo Zero is a closed model — its weights were never published, so it cannot be downloaded or run on your own hardware at any price. It is reachable only through its provider.

03

Is AlphaGo Zero open source?

No. AlphaGo Zero has not had its weights published, so it exists only as a service controlled by its owner.

04

How many parameters does AlphaGo Zero have?

AlphaGo Zero has 46.4M parameters. Input size: 19*19*17=6137, internal dimension 19*19 1 conv block 17*3*3*256=39168 39 residual blocks 2*3*3*256*256=1179648 Policy head 256*2+2*19*19*(19*19+2)=262598 Value head 256*1+1*19*19*256+256*1=92928 Total: 39168+39*1179648+262598+92928=46400966. That figure is the total, and it is what decides how much memory the model needs — roughly half a gigabyte per billion at the compression most people use.

05

Who created AlphaGo Zero?

AlphaGo Zero was published by DeepMind, based in United Kingdom of Great Britain and Northern Ireland, categorised as industry.

06

When was AlphaGo Zero released?

AlphaGo Zero was published in October 2017. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.

07

What is AlphaGo Zero used for?

AlphaGo Zero works in Games, and is recorded as handling go. A model can carry several of each, so these are the areas it was built for rather than a limit on what it will attempt.

Source

Original publication

Record last updated 1 January 2026

The other direction

Looking at it from the other side?

This page starts from the model. If you already own a card and want to know everything it will run, start from the hardware instead.