Skip to content

fix(providers): auto-inflate max_tokens for reasoning models #379

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Bahtya merged 1 commit into from
May 14, 2026
Merged

fix(providers): auto-inflate max_tokens for reasoning models #379

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
20 changes: 19 additions & 1 deletion crates/kestrel-providers/src/openai_compat.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -136,7 +136,12 @@ impl OpenAiCompatProvider {
});

if let Some(max_tokens) = request.max_tokens {
body["max_tokens"] = json!(max_tokens);
let effective = if is_reasoning_model(&request.model) {
max_tokens.max(16384)
} else {
max_tokens
};
body["max_tokens"] = json!(effective);
}
if let Some(temp) = request.temperature {
body["temperature"] = json!(temp);
Expand Down Expand Up @@ -381,6 +386,19 @@ fn build_openai_tool_call_deltas(
Some(deltas)
}

/// Check if a model is a known reasoning model that uses thinking tokens.
///
/// Reasoning models consume `max_tokens` budget with thinking/reasoning tokens.
/// We auto-inflate the budget so thinking doesn't crowd out actual content.
fn is_reasoning_model(model: &str) -> bool {
let m = model.to_lowercase();
m.contains("glm-5")
|| m.starts_with("deepseek-r")
|| m.starts_with("o1")
|| m.starts_with("o3")
|| m.starts_with("o4")
}

#[derive(Debug, Deserialize)]
struct OpenAiResponse {
choices: Vec<OpenAiChoice>,
Expand Down
Loading