The LLM API Pricing Index

Methodology & sources

The StackSpend LLM Pricing Index is refreshed daily. Prices are provider list prices in USD per 1 million tokens, synced from the community-maintained LiteLLM price dataset. Coding scores come from the Aider polyglot benchmark and are attached to the underlying base model, so the identical open-weight model served by different hosts shares one score. StackSpend uses this same data to track and right-size your own AI spend. For what these models are and how the families relate, see the LLM model glossary.

How often is this LLM pricing data updated?

Daily. Prices are synced from the community-maintained LiteLLM price dataset and coding scores from the Aider polyglot benchmark; this page was last refreshed on 11 July 2026. All prices are list prices in USD per 1M tokens.

Which model gives the best coding performance for the price?

Among models benchmarked for coding, deepseek-v3.2 (DeepSeek) offers the strongest coding score relative to its output price — 74.2% on the Aider polyglot benchmark at $0.40 per 1M output tokens.

What is the highest-scoring model for coding?

gemini-2.5-pro (Google (Gemini)) currently leads the coding benchmark at 83.1%, priced at $1 input / $10 output per 1M tokens.

Why do the same open models cost different amounts?

Open-weight models (like Llama or DeepSeek) are served by multiple inference hosts — Groq, Together AI, Fireworks, DeepInfra and others — at different prices for the identical weights. Comparing hosts for the same model is often the fastest saving.