Skip to content

Use Claude Code with your ChatGPT, Kimi, Cursor or Grok subscription via a local Anthropic-compatible proxy

Rust

cap codex output tokens

Raine Virta607d014b5e72
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 2cfd099..309c502 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -11,6 +11,8 @@
- Codex WebSocket live streaming reuses pooled session sockets and applies
`previous_response_id` continuations on append-only turns, reducing repeated
upstream input for streaming sessions.
+- Codex requests forward Claude Code's `max_tokens` as `max_output_tokens`,
+ keeping upstream responses within Claude Code's configured output limit.
## v0.1.2 (2026-06-30)
diff --git a/src/providers/codex/client.rs b/src/providers/codex/client.rs
index 6ed2c04..899e736 100644
--- a/src/providers/codex/client.rs
+++ b/src/providers/codex/client.rs
@@ -948,6 +948,7 @@ mod tests {
];
let req = ResponsesRequest {
model: "gpt-5.5".to_string(),
+ max_output_tokens: 32_000,
instructions: None,
input,
tools: None,
@@ -969,6 +970,7 @@ mod tests {
payload.get("type").and_then(|v| v.as_str()),
Some("response.create")
);
+ assert_eq!(payload["max_output_tokens"], 32_000);
assert!(payload.get("stream").is_none());
assert!(payload.get("previous_response_id").is_none());
}
diff --git a/src/providers/codex/request_summary.rs b/src/providers/codex/request_summary.rs
index 39906dd..88d3a5c 100644
--- a/src/providers/codex/request_summary.rs
+++ b/src/providers/codex/request_summary.rs
@@ -229,6 +229,7 @@ mod tests {
}];
let req = ResponsesRequest {
model: "gpt-5.5".to_string(),
+ max_output_tokens: 32_000,
instructions: None,
input,
tools: None,
diff --git a/src/providers/codex/translate/request.rs b/src/providers/codex/translate/request.rs
index 282683f..e65b6f0 100644
--- a/src/providers/codex/translate/request.rs
+++ b/src/providers/codex/translate/request.rs
@@ -53,6 +53,8 @@ pub enum ResponsesToolChoice {
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ResponsesRequest {
pub model: String,
+ #[serde(default = "default_max_output_tokens")]
+ pub max_output_tokens: u32,
#[serde(default, skip_serializing_if = "Option::is_none")]
pub instructions: Option<String>,
pub input: Vec<ResponsesInputItem>,
@@ -183,6 +185,8 @@ pub struct TranslateOptions {
pub model: String,
}
+const DEFAULT_MAX_OUTPUT_TOKENS: u32 = 32_000;
+
// ---------------------------------------------------------------------------
// Translation entry point
// ---------------------------------------------------------------------------
@@ -287,6 +291,7 @@ pub fn translate_request(
let mut out = ResponsesRequest {
model: opts.model,
+ max_output_tokens: clamp_max_output_tokens(req.max_tokens),
instructions,
input,
store: false,
@@ -339,6 +344,17 @@ pub fn translate_request(
// Helpers
// ---------------------------------------------------------------------------
+fn default_max_output_tokens() -> u32 {
+ DEFAULT_MAX_OUTPUT_TOKENS
+}
+
+fn clamp_max_output_tokens(requested: Option<u32>) -> u32 {
+ match requested {
+ Some(value) if value > 0 => value.min(DEFAULT_MAX_OUTPUT_TOKENS),
+ _ => DEFAULT_MAX_OUTPUT_TOKENS,
+ }
+}
+
fn read_output_format(req: &MessagesRequest) -> Option<ResponsesTextFormat> {
let output_config = req.extra.get("output_config")?.as_object()?;
let format = output_config.get("format")?.as_object()?;
@@ -707,6 +723,41 @@ mod tests {
assert!(matches!(out.reasoning.unwrap().effort, Some(Effort::Xhigh)));
}
+ #[test]
+ fn max_output_tokens_defaults_to_32000() {
+ let req: MessagesRequest = serde_json::from_value(json!({
+ "model": "gpt-5.5",
+ "messages": [{"role":"user", "content":"hello"}]
+ }))
+ .unwrap();
+ let out = translate_request(&req, opts()).unwrap();
+ assert_eq!(out.max_output_tokens, 32_000);
+ }
+
+ #[test]
+ fn max_output_tokens_uses_request_limit() {
+ let req: MessagesRequest = serde_json::from_value(json!({
+ "model": "gpt-5.5",
+ "max_tokens": 4096,
+ "messages": [{"role":"user", "content":"hello"}]
+ }))
+ .unwrap();
+ let out = translate_request(&req, opts()).unwrap();
+ assert_eq!(out.max_output_tokens, 4096);
+ }
+
+ #[test]
+ fn max_output_tokens_clamps_at_32000() {
+ let req: MessagesRequest = serde_json::from_value(json!({
+ "model": "gpt-5.5",
+ "max_tokens": 99999,
+ "messages": [{"role":"user", "content":"hello"}]
+ }))
+ .unwrap();
+ let out = translate_request(&req, opts()).unwrap();
+ assert_eq!(out.max_output_tokens, 32_000);
+ }
+
#[test]
fn reasoning_summary_override_values() {
assert!(reasoning_summary_requested(None));
diff --git a/tests/smoke_cutover.rs b/tests/smoke_cutover.rs
index b9b6451..fe5d69f 100644
--- a/tests/smoke_cutover.rs
+++ b/tests/smoke_cutover.rs
@@ -756,6 +756,7 @@ async fn smoke_codex_websocket_messages_uses_mock_upstream() {
});
assert_eq!(sent["type"], "response.create");
assert_eq!(sent["model"], "gpt-5.5");
+ assert_eq!(sent["max_output_tokens"], 64);
assert!(sent.get("stream").is_none());
}