Skip to content
← Models
ClosedGoogle DeepMind/10 trustMetrics-checked

AlphaCode

Code-generation system that reached median human level in competitive programming.

Visit official page →

About

AlphaCode is DeepMind's code generation system for solving competitive programming problems, unveiled in February 2022 with a paper in Science. It uses large transformer language models pretrained on GitHub code and fine-tuned on competitive programming problems, generating large pools of candidate programs that are filtered by execution on example tests and clustered into a small set of submissions.

In simulated evaluations on ten recent Codeforces contests with 5,000+ participants each, AlphaCode ranked on average in the top 54.3%, roughly the level of a median human competitor, with an estimated Codeforces rating of 1238 (top 28% of active users). It was the first AI system to reach a competitive level in programming contests. It is a research system rather than a product: no model weights were publicly released, and the program was later succeeded by AlphaCode 2 (Gemini-powered, 2023).

Specs

Context window

Pricing

Free quota

API access

Weights

Closed — API only

Released

February 2, 2022

Benchmarks

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

code-generationcompetitive-programmingresearch

Trust score — 7/10

/10 trustMetrics-checked

Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 84; Trending score 74/100 from researcher assessment.

Sources