More complete cost information

[cpsievert]

litellm maintains a list here https://github.com/BerriAI/litellm/blob/main/model_prices_and_context_window.json

[hadley]

Looks nice, thanks!

[hadley]

library(dplyr)
library(tidyr)

litellm_url <- "https://raw.githubusercontent.com/BerriAI/litellm/refs/heads/main/model_prices_and_context_window.json"
litellm_prices <- jsonlite::read_json(litellm_url)

# fmt: skip
provider_lookup <- tribble(
  ~litellm_provider, ~provider, 
  "openai", "OpenAI",
  "anthropic", "Anthropic",
  "gemini", "Google/Gemini"
)

df <- tibble::enframe(litellm_prices, "model", "data")
df |>
  hoist(
    data,
    provider = "litellm_provider",
    input = "input_cost_per_token",
    output = "output_cost_per_token",
    cached_input = "cache_read_input_token_cost"
  ) |>
  filter(model != "sample_spec") |>
  mutate(
    input = input * 1e6,
    output = output * 1e6,
    cached_input = cached_input * 1e6,
    data = NULL,
    model = stringr::str_remove(model, paste0(provider, "/"))
  ) |>
  relocate(provider) |>
  arrange(provider, model) |>
  filter(!is.na(input) | !is.na(output)) |>
  inner_join(provider_lookup, join_by(provider == litellm_provider)) |> 
  mutate(provider = provider.y, provider.y = NULL) |>
  View()

[hadley]

Some prices are definitely wrong, e.g.

    "o1-mini-2024-09-12": {
        "max_tokens": 65536,
        "max_input_tokens": 128000,
        "max_output_tokens": 65536,
        "input_cost_per_token": 3e-06,
        "output_cost_per_token": 1.2e-05,
        "cache_read_input_token_cost": 1.5e-06,
        "litellm_provider": "openai",
        "mode": "chat",
        "supports_pdf_input": true,
        "supports_vision": true,
        "supports_reasoning": true,
        "supports_prompt_caching": true
    }

But input cost per million tokens per https://platform.openai.com/docs/pricing is $1.10.