Problem

I'm using ellmer with my company's internal OpenAI endpoint.

This worked in ellmer v0.3.0:

library(ellmer)
chat <- chat_openai(
  model = "bedrock-anthropic-claude-3.7-sonnet",
  base_url = "internal-endpoint",
)
chat$chat("What is the difference between a tibble and a data frame?")

In ellmer v0.4.0, I'd start to get a response, then:

Error in data.frame(provider = provider@name, model = provider@model,
  arguments imply differing number of rows: 1, 0

Likely Cause

get_token_cost() is called with variant = NULL when the API response doesn't include service_tier. The data.frame() call then fails because variant has length 0 while other columns have length 1.

Potential Fix

Default variant to NULL and set to "" if so:

get_token_cost <- function(provider, tokens, variant = NULL) {
  variant <- variant %||% ""
  # ...
}