fix(cost): match Python's Vertex character routing and Fireworks size parsing

Vertex routes and prices characters with the provider prefix stripped, so
prefixed gemini-pro keeps its fixed pricing, and Fireworks parses the
parameter count as a float so oversized counts land in the above-16b tier

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Yujong Lee 2026-09-23 19:16:12 -07:00
parent 76decc8ae7
commit bf6ea4d781
4 changed files with 51 additions and 5 deletions

View file

@ -56,12 +56,12 @@ pub fn get_base_model_for_pricing(
}
if let Some(size) = MODEL_SIZE
.captures(&name)
.and_then(|captures| captures.get(1)?.as_str().parse::<u64>().ok())
.and_then(|captures| captures.get(1)?.as_str().parse::<f64>().ok())
{
if size <= thresholds.small {
if size <= thresholds.small as f64 {
return "fireworks-ai-up-to-4b";
}
if size <= thresholds.medium {
if size <= thresholds.medium as f64 {
return "fireworks-ai-4.1b-to-16b";
}
return "fireworks-ai-above-16b";

View file

@ -140,7 +140,15 @@ pub fn vertex_cost(
{
return Ok(cost);
}
if cost_router(request.model, request.provider.unwrap_or(""), call_type) == CostRoute::PerToken
let model_without_prefix = request
.model
.split_once('/')
.map_or(request.model, |(_, model)| model);
if cost_router(
model_without_prefix,
request.provider.unwrap_or(""),
call_type,
) == CostRoute::PerToken
{
return catalog_cost_per_token(catalog, request);
}
@ -148,7 +156,7 @@ pub fn vertex_cost(
.entry(request.model, request.provider, request.region)
.ok_or(CostError::ModelNotFound)?;
Ok(cost_per_character(
request.model,
model_without_prefix,
request.usage,
entry,
(prompt_characters, completion_characters),

View file

@ -29,6 +29,8 @@ fn usage(value: Value) -> litellm_cost::responses_usage::ChatUsage {
#[case("model-8x22b", "fireworks-ai-56b-to-176b")]
#[case("model-8x30b", "fireworks-ai-above-16b")]
#[case("model-without-size", "fireworks-ai-default")]
#[case("foo-99999999999999999999b", "fireworks-ai-above-16b")]
#[case("99999999999999999999x2b", "fireworks-ai-up-to-4b")]
fn fireworks_model_size_category_matches_python_pattern_order(
#[case] model: &str,
#[case] expected: &str,

View file

@ -182,3 +182,39 @@ fn catalog_vertex_cost_routes_token_and_character_calls() {
(0.01, 0.01)
);
}
#[rstest]
#[case::bare_model("gemini-pro")]
#[case::provider_prefixed_model("vertex_ai/gemini-pro")]
fn vertex_cost_matches_models_without_dynamic_pricing_after_stripping_the_prefix(
#[case] model: &str,
) {
let catalog = ModelInfoCatalog::new(HashMap::from([(
"vertex_ai/gemini-pro".to_owned(),
json!({
"input_cost_per_character": 1e-6,
"input_cost_per_character_above_128k_tokens": 9e-6,
"output_cost_per_character": 1e-6
}),
)]));
let usage = ChatUsage::default();
let (prompt, _) = vertex_cost(
&catalog,
ModelCostRequest {
model,
provider: Some("vertex_ai"),
region: None,
usage: &usage,
service_tier: None,
data_residency: None,
vertex_location: None,
at: at(),
response_time_ms: None,
},
"completion",
Some(40_000.0),
Some(0.0),
)
.unwrap();
assert!((prompt - 0.04).abs() < 1e-12);
}