Note
This article is about prices as of January 11, 2024. For current prices and more comprehensive analysis, check artificialanalysis.ai (not affiliated with me).
This is an overview of pricing for large language models from different developers and API providers. The dataset is available on GitHub. Prices are expressed in USD per 1 million tokens. To learn more about tokens, see the Tokenizer by OpenAI.
Price comparison
Hover over bars to see extra information (also available in table below). The prices for input and output tokens were averaged. For AWS, the region us-east-1 was used.
- Price differences are huge, with a 600x difference between the cheapest and most expensive models ($0.15 vs $90)
- GPT-4 is the most expensive model, followed by GPT-3.5 and PaLM2
- Prices on Azure and OpenAI are identical
- Anyscale is the cheapest provider for large models, serving Mistral’s models at lower prices than Mistral itself
- Prices roughly reflect the number of parameters in the models, which again roughly map to their capability
Papers with Code has a leaderboard for the MMLU (Massive Multitask Language Understanding) benchmark. The HuggingFace OpenLLM Leaderboard offers a more detailed ranking of open source models across different benchmarks. These leaderboards don’t have benchmarks for every model listed here.
Model table
Click on column headers to sort. On mobile, scroll right to see all columns.
| Model | Provider | Developer | Context size | Input $/1M | Output $/1M | Avg. $/1M |
|---|---|---|---|---|---|---|
| Loading... (need help?) |