OpenAI Five
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
- OpenAI
- Organisation type
- Industry
- Country
- United States of America
- Published
- 13 December 2019
- Authors
- Christopher Berner, Greg Brockman, Brooke Chan, Vicki Cheung, Przemysław Dębiak, Christy Dennison, David Farhi, Quirin Fischer, Shariq Hashme, Chris Hesse, Rafal Józefowicz, Scott Gray, Catherine Olsson, Jakub Pachocki, Michael Petrov, Henrique P. d.O. Pinto, Jonathan Raiman, Tim Salimans, Jeremy Schlatter, Jonas Schneider, Szymon Sidor, Ilya Sutskever, Jie Tang, Filip Wolski, Susan Zhang
What it does
The problem areas the model was built for. A model can carry several of each.
- Domain
- Games
- Task
- Dota 2
- 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
- 159M
- Training data
- 454,164,480,000 tokens
"We define a policy (π) as a function from the history of observations to a probability distribution over actions, which we parameterize as a recurrent neural network with approximately 159 million parameters (θ)." pg. 3 of paper source: https://docs.google.com/spreadsheets/d/1Kj4Q5WADcDXtUJLIOfGTCE3tGvxNczEMwyy8QtgSkHk/edit#gid=54587040&fvid=1361937389
"Although the Dota 2 engine runs at 30 frames per second, OpenAI Five only acts on every 4th frame which we call a timestep" --> 7.5 timesteps/s "OpenAI Five is a single training run that ran from June 30th, 2018 to April 22nd, 2019. " --> 296 days 296 * 24*3600 * 7.5 = 1.92e8 This number seems a little low? The DQN paper had 1e7 timesteps. Might be to do with sample efficiency? EDIT 14/06/2022 Multiple copies of OpenAI Five were trained in parallel, so the total training time is much higher…
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
- 6.7 × 10²² FLOP
- How it was established
- Reported
"770±50 PFlops/s·days of compute" for the model that played against world champions. They did a single training run that took 10 months. While the model was playing against world champions, they continued training for a few days, so that the resulting model used even more training compute: 820±50 PFlops/s·days. Finally, they also trained a Rerun model with 150±5 PFlops/s·days of compute. Source: Dota 2 with Large Scale Deep Reinforcement Learning https://arxiv.org/abs/1912.06680 You cannot m…
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
- NVIDIA P100
- Chips used
- 1,536
- Chip-hours
- 10,911,744
- Wall-clock time
- 7,104 hours (296 days)
- Power draw
- 786.4 kW
- Compute cost
- $4,328,311
"OpenAI Five is a single training run that ran from June 30th, 2018 to April 22nd, 2019. " --> 296 days
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,SOTA improvement
- Record confidence
- Confident
- Citations
- 2,137
no standard benchmark results "On April 13th, 2019, OpenAI Five became the first AI system to defeat the world champions at an esports game." " it defeated the Dota 2 world champions in a best-of-three match and 99.4% of human players during a multi-day online showcase." "we ran OpenAI Five Arena, in which we opened OpenAI Five to the public for competitive online games from April 18-21, 2019. In total, Five played 3,193 teams in 7,257 total games, winning 99.4%"
Sources
Where this record came from and when it was last checked.
- Reference
- Dota 2 with Large Scale Deep Reinforcement Learning
- Last updated
- 25 May 2026
What the numbers mean
What this model is
OpenAI Five was published by OpenAI, in United States of America, in December 2019. The organisation is categorised as industry.
It works in Games, and is recorded as doing dota 2.
This is a closed model: the trained values stayed with whoever produced them, and there is no local version to run.
What went into building it
The training run consumed about 6.7 × 10²² FLOP, on NVIDIA P100. That figure describes the cost of creating it and has no bearing on how quickly it generates text.
The training set ran to roughly 454,164,480,000 tokens.
It is tracked in the underlying dataset for one reason in particular: highly cited,SOTA improvement.
Answers
OpenAI Five — common questions
Who created OpenAI Five?
OpenAI Five was published by OpenAI, based in United States of America, categorised as industry.
When was OpenAI Five released?
OpenAI Five was published in December 2019. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.
What is OpenAI Five used for?
OpenAI Five works in Games, and is recorded as handling dota 2. These are the areas it was designed around; they describe intent rather than a hard boundary.
How much compute was used to train OpenAI Five?
Around 6.7 × 10²² FLOP, on NVIDIA P100. 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.
What GPU do I need to run OpenAI Five?
None. OpenAI Five 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.
Is OpenAI Five open source?
No. OpenAI Five has not had its weights published, so it exists only as a service controlled by its owner.
How many parameters does OpenAI Five have?
OpenAI Five has 159M parameters. "We define a policy (π) as a function from the history of observations to a probability distribution over actions, which we parameterize as a recurrent neural network with approximately 159 million parameters (θ)." pg. 3 of paper source: https://docs.google.com/spreadsheets/d/1Kj4Q5WADcDXtUJLIOfGTCE3tGvxNczEMwyy8QtgSkHk/edit#gid=54587040&fvid=1361937389. 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.
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.