AlphaCode
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
- 2 February 2022
- Authors
- Yujia Li, David Choi, Junyoung Chung, Nate Kushman, Julian Schrittwieser, Rémi Leblond, Tom Eccles, James Keeling, Felix Gimeno, Agustin Dal Lago, Thomas Hubert, Peter Choy, Cyprien de Masson d'Autume, Igor Babuschkin, Xinyun Chen, Po-Sen Huang, Johannes Welbl, Sven Gowal, Alexey Cherepanov, James Molloy, Daniel J. Mankowitz, Esme Sutherland Robson, Pushmeet Kohli, Nando de Freitas, Koray Kavukcuo…
What it does
The problem areas the model was built for. A model can carry several of each.
- Domain
- Language
- Task
- Code generation
- Approach
- Self-supervised learning
- Numerical format
- BF16
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
- 41.1B
- Training data
- 967,000,000,000 tokens
- Batch size
- 4,718,592
41.1B. Table 3
Appendix part A has answers for pretraining.
2304 token sequences, 2048 batch size. 2304 * 2048 = 4718592 trained on 967B tokens and 205k steps. 967B/205k = 4717073, so seems they didn't do warmup
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
- 2.4 × 10²³ FLOP
- How it was established
- Hardware
Using C=6ND, we have C = 6 FLOP/token/param * 41.1B params * 967B tokens = 2.38e23 FLOP. Figure 7 (a) shows a maximum training compute budget of approx 23000 TPU-days per model. This matches the operation-counting estimate at 44% utilization.
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 v4
- Chips used
- 3,750
- Wall-clock time
- 147 hours
- Hardware utilisation
- MFU 43.6%
- Power draw
- 2.6 MW
Figure 7 (a) shows that the models were trained for around 23000 TPU-days. We know they trained on TPUv4s, and in appendix D.1 they say they have 3750 TPUv4 and TPUv4i. Assuming they trained only on the 3750 TPUv4s, that suggests 23000 / 3750 = 6.13 days, or 147.2 hours.
Using C=6ND, we have C = 6 FLOP/token/param * 41.1B params * 967B tokens = 2.38e23 FLOP. Figure 7 (a) shows a maximum training compute budget of approx 23000 TPU-days per model. This implies 2.38e23 / (275e12 * 86400 * 23000) = 43.64% MFU = 0.4364
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.
- Foundation model
- Yes
- Likely above 10²³ FLOP
- Yes
- Why it is tracked
- SOTA improvement
- Record confidence
- Confident
- Citations
- 2,162
" To the best of our knowledge, this is the first time that a computer system has been competitive with human participants in programming competitions"
Sources
Where this record came from and when it was last checked.
- Reference
- Competition-Level Code Generation with AlphaCode
- Last updated
- 25 May 2026
What the numbers mean
Where it came from
AlphaCode was published by DeepMind, in United Kingdom of Great Britain and Northern Ireland, in February 2022. It comes out of industry.
It works in Language, and is recorded as doing code generation.
Because the weights are not available, none of the hardware figures elsewhere on this site apply to it.
How it was trained
Producing it required around 2.4 × 10²³ FLOP of arithmetic, on Google TPU v4, which is a statement about the training budget rather than about inference.
The training set ran to roughly 967,000,000,000 tokens.
Its inclusion criterion is sOTA improvement.
Answers
AlphaCode — common questions
What GPU do I need to run AlphaCode?
None. AlphaCode 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 AlphaCode open source?
No. AlphaCode has not had its weights published, so it exists only as a service controlled by its owner.
How many parameters does AlphaCode have?
AlphaCode has 41.1B parameters. 41.1B. Table 3. 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.
Who created AlphaCode?
AlphaCode was published by DeepMind, based in United Kingdom of Great Britain and Northern Ireland, categorised as industry.
When was AlphaCode released?
AlphaCode was published in February 2022. 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 AlphaCode used for?
AlphaCode works in Language, and is recorded as handling code generation. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.
How much compute was used to train AlphaCode?
Around 2.4 × 10²³ FLOP, on Google TPU v4. 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.
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.