Skip to content

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

Rust

omit unsupported codex output cap

Raine Virta9a52d5461926
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 309c502..8ade092 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -11,8 +11,9 @@
- 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.
+- Codex WebSocket requests use the gateway-compatible request shape, so
+ `claude -p` jobs and release changelog generation avoid unsupported-parameter
+ failures.
## v0.1.2 (2026-06-30)
diff --git a/src/providers/codex/client.rs b/src/providers/codex/client.rs
index 899e736..6ed2c04 100644
--- a/src/providers/codex/client.rs
+++ b/src/providers/codex/client.rs
@@ -948,7 +948,6 @@ mod tests {
];
let req = ResponsesRequest {
model: "gpt-5.5".to_string(),
- max_output_tokens: 32_000,
instructions: None,
input,
tools: None,
@@ -970,7 +969,6 @@ 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 88d3a5c..39906dd 100644
--- a/src/providers/codex/request_summary.rs
+++ b/src/providers/codex/request_summary.rs
@@ -229,7 +229,6 @@ 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 e65b6f0..e7c1e67 100644
--- a/src/providers/codex/translate/request.rs
+++ b/src/providers/codex/translate/request.rs
@@ -53,8 +53,6 @@ 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>,
@@ -185,8 +183,6 @@ pub struct TranslateOptions {
pub model: String,
}
-const DEFAULT_MAX_OUTPUT_TOKENS: u32 = 32_000;
-
// ---------------------------------------------------------------------------
// Translation entry point
// ---------------------------------------------------------------------------
@@ -291,7 +287,6 @@ 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,
@@ -344,17 +339,6 @@ 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()?;
@@ -724,18 +708,7 @@ mod tests {
}
#[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() {
+ fn max_tokens_is_not_serialized_for_codex() {
let req: MessagesRequest = serde_json::from_value(json!({
"model": "gpt-5.5",
"max_tokens": 4096,
@@ -743,19 +716,8 @@ mod tests {
}))
.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);
+ let value = serde_json::to_value(out).unwrap();
+ assert!(value.get("max_output_tokens").is_none());
}
#[test]
diff --git a/tests/smoke_cutover.rs b/tests/smoke_cutover.rs
index fe5d69f..bdc5bbd 100644
--- a/tests/smoke_cutover.rs
+++ b/tests/smoke_cutover.rs
@@ -756,7 +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("max_output_tokens").is_none());
assert!(sent.get("stream").is_none());
}