GPT-2
Language Models are Unsupervised Multitask Learners
OpenAI's 1.5 billion parameter language model — about 13× larger than its predecessor — that could complete arbitrary text prompts fluently enough that OpenAI initially refused to release the full weights, citing misuse risk. Six months later, they released them anyway when the predicted misuse failed to materialize at scale.
Cost
Free
Open weights — self-host
Context
1K
Up to 1,024 tokens
How are Intelligence, Speed & Cost bucketed?
Intelligence and Speed buckets are percentile ranks on
Artificial Analysis. Cost buckets are fixed dollar
thresholds keyed off output-token price ($/M out).
Intelligence
- Top 1%≤ 1%
- Top 5%≤ 5%
- Top 10%≤ 10%
- Good≤ 25%
- Medium≤ 50%
- Below avg> 50%
Speed
- Top 1%≥ 345 tok/s
- Top 5%≥ 237 tok/s
- Top 10%≥ 196 tok/s
- Good≥ 146 tok/s
- Medium≥ 90 tok/s
- Slow< 90 tok/s
Cost
- Freeopen weights · self-host
- Low< $1 / M out
- Moderate$1–5 / M out
- High≥ $5 / M out
Why it matters
The first language model whose outputs were good enough to fool casual readers. Set the template for staged release controversies and made "this model is too dangerous to release" a meaningful claim that companies, regulators, and journalists would all subsequently take seriously.
Core Capabilities
Long Documents
Handles entire codebases, books, and multi-doc RAG.
Generative
Produces images, video, audio, or other media.
Context Window
1k tokens
short prompt
4k Chat 聊天
32k Long docs 长文档
128k Books 整本书
400k Multi-doc 多文档
1M Codebase 整个代码库
10M
1k
Availability
API
Not available
Product / App
Not available
Open Source
Released
Enterprise
—
Pricing Model
Free / self-host
Open weights — pay only for compute.
Self-host Capability / Performance
Where this model sits relative to the middle 60% of models in the tree. All scores are 0–10 (higher is better).
Lower 20% Upper 80% This model
Context / memory
Context window size · log-scaled
6.0
9.0
0.0
Lower 20% 20th percentile — 20% of models score below this This model Where the current model lands Upper 80% 80th percentile — only 20% of models score above this
Percentile boundaries are computed across every model in the tree that reports the underlying benchmark for each capability.
What it feels like
- Original GPT-2 — historical; preceded ChatGPT.
- Step-change in reasoning vs GPT-3.5 — top 10% on simulated bar exam vs 3.5's bottom 10%
- MMLU 86.4% in English; surpassed prior models in 24 of 26 other languages
- First widely-deployed model with image input (text output only) — multimodal era starts here
Best use cases
- Professional knowledge work needing top-of-class reasoning at the time
- Code generation with chain-of-thought prompting
- Multilingual tasks across 26+ languages
Tools to try
Not ideal for
- Frontier work after Claude 3.5+ / GPT-4o / Llama 3 — quickly surpassed in 2024
- Cost-sensitive bulk inference — pricing dominant before GPT-4o cut it in half