The cost of LLMs is steadily falling, and their quality is rising. This chart helps you navigate the trade-offs.
We use the ELO score (without style control) from the LM Arena leaderboard as a proxy for quality. It's like a chess Elo rating, but for LLMs, based on thousands of human votes.
We use the cost per million input tokens as a proxy for cost, based on data from the LLM Pricing Calculator. Output tokens are typically more expensive, but input tokens usually make up the bulk of the cost.