mirror of
https://github.com/openai/codex.git
synced 2026-11-18 23:02:23 +00:00
Support persistent reasoning effort (#40799)
## What changed - Add `persistent` to the reasoning-effort protocol and TypeScript SDK types. - Show model-advertised persistent reasoning in the TUI as “Persistent”. - Preserve `persistent` in local configuration while sending the Responses API's `disabled` wire value. ## Testing - Cover parsing and serialization, request translation, remote model requests, the TUI reasoning selector, and TypeScript CLI argument forwarding. GitOrigin-RevId: 22fc2fba1975db345c12d555ec12803d8dc0abca
This commit is contained in:
1 parent
37f4bb94c9
commit
3e4707b34b
9 files changed
+72
-39
No files matched your search
@@ -180,6 +180,8 @@ pub(crate) struct CompactConversationRequestSettings {
|
|||||||
fn reasoning_effort_for_request(effort: ReasoningEffortConfig) -> ReasoningEffortConfig {
|
fn reasoning_effort_for_request(effort: ReasoningEffortConfig) -> ReasoningEffortConfig {
|
||||||
match effort {
|
match effort {
|
||||||
ReasoningEffortConfig::Ultra => ReasoningEffortConfig::Max,
|
ReasoningEffortConfig::Ultra => ReasoningEffortConfig::Max,
|
||||||
|
// Keep "persistent" in local settings; the Responses API calls it "disabled".
|
||||||
|
ReasoningEffortConfig::Persistent => ReasoningEffortConfig::Custom("disabled".to_string()),
|
||||||
effort => effort,
|
effort => effort,
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|||||||
@@ -299,13 +299,18 @@ fn test_session_telemetry() -> SessionTelemetry {
|
|||||||
}
|
}
|
||||||
|
|
||||||
#[test]
|
#[test]
|
||||||
fn ultra_reasoning_uses_max_for_requests() {
|
fn reasoning_effort_for_requests_maps_ultra_and_persistent() {
|
||||||
assert_eq!(
|
assert_eq!(
|
||||||
(
|
(
|
||||||
super::reasoning_effort_for_request(ReasoningEffort::Ultra),
|
super::reasoning_effort_for_request(ReasoningEffort::Ultra),
|
||||||
super::reasoning_effort_for_request(ReasoningEffort::High),
|
super::reasoning_effort_for_request(ReasoningEffort::High),
|
||||||
|
super::reasoning_effort_for_request(ReasoningEffort::Persistent),
|
||||||
),
|
),
|
||||||
(ReasoningEffort::Max, ReasoningEffort::High,)
|
(
|
||||||
|
ReasoningEffort::Max,
|
||||||
|
ReasoningEffort::High,
|
||||||
|
ReasoningEffort::Custom("disabled".to_string()),
|
||||||
|
)
|
||||||
);
|
);
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|||||||
@@ -48,6 +48,7 @@ use core_test_support::wait_for_event_match;
|
|||||||
use pretty_assertions::assert_eq;
|
use pretty_assertions::assert_eq;
|
||||||
use serde_json::json;
|
use serde_json::json;
|
||||||
use tempfile::TempDir;
|
use tempfile::TempDir;
|
||||||
|
use test_case::test_case;
|
||||||
use tokio::time::Duration;
|
use tokio::time::Duration;
|
||||||
use tokio::time::Instant;
|
use tokio::time::Instant;
|
||||||
use tokio::time::sleep;
|
use tokio::time::sleep;
|
||||||
@@ -365,8 +366,13 @@ async fn remote_models_use_context_window_when_config_override_is_absent() -> Re
|
|||||||
Ok(())
|
Ok(())
|
||||||
}
|
}
|
||||||
|
|
||||||
|
#[test_case(ReasoningEffort::Custom("future".to_string()), "future"; "custom")]
|
||||||
|
#[test_case(ReasoningEffort::Persistent, "disabled"; "persistent")]
|
||||||
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
|
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
|
||||||
async fn remote_models_long_model_slug_is_sent_with_custom_reasoning() -> Result<()> {
|
async fn remote_models_long_model_slug_is_sent_with_supported_reasoning(
|
||||||
|
effort: ReasoningEffort,
|
||||||
|
expected_wire_effort: &str,
|
||||||
|
) -> Result<()> {
|
||||||
skip_if_no_network!(Ok(()));
|
skip_if_no_network!(Ok(()));
|
||||||
skip_if_sandbox!(Ok(()));
|
skip_if_sandbox!(Ok(()));
|
||||||
|
|
||||||
@@ -379,16 +385,15 @@ async fn remote_models_long_model_slug_is_sent_with_custom_reasoning() -> Result
|
|||||||
/*priority*/ 1_000,
|
/*priority*/ 1_000,
|
||||||
TruncationPolicyConfig::bytes(/*limit*/ 10_000),
|
TruncationPolicyConfig::bytes(/*limit*/ 10_000),
|
||||||
);
|
);
|
||||||
let custom_reasoning_effort = ReasoningEffort::Custom("future".to_string());
|
remote_model.default_reasoning_level = Some(effort.clone());
|
||||||
remote_model.default_reasoning_level = Some(custom_reasoning_effort.clone());
|
|
||||||
remote_model.supported_reasoning_levels = vec![
|
remote_model.supported_reasoning_levels = vec![
|
||||||
ReasoningEffortPreset {
|
ReasoningEffortPreset {
|
||||||
effort: ReasoningEffort::Medium,
|
effort: ReasoningEffort::Medium,
|
||||||
description: ReasoningEffort::Medium.to_string(),
|
description: ReasoningEffort::Medium.to_string(),
|
||||||
},
|
},
|
||||||
ReasoningEffortPreset {
|
ReasoningEffortPreset {
|
||||||
effort: custom_reasoning_effort.clone(),
|
effort: effort.clone(),
|
||||||
description: custom_reasoning_effort.to_string(),
|
description: effort.to_string(),
|
||||||
},
|
},
|
||||||
];
|
];
|
||||||
remote_model.default_reasoning_summary = ReasoningSummary::Detailed;
|
remote_model.default_reasoning_summary = ReasoningSummary::Detailed;
|
||||||
@@ -411,7 +416,7 @@ async fn remote_models_long_model_slug_is_sent_with_custom_reasoning() -> Result
|
|||||||
.with_config(|config| {
|
.with_config(|config| {
|
||||||
config.model = Some(requested_model.to_string());
|
config.model = Some(requested_model.to_string());
|
||||||
})
|
})
|
||||||
.build(&server)
|
.build_with_auto_env(&server)
|
||||||
.await?;
|
.await?;
|
||||||
|
|
||||||
codex
|
codex
|
||||||
@@ -434,7 +439,7 @@ async fn remote_models_long_model_slug_is_sent_with_custom_reasoning() -> Result
|
|||||||
.and_then(|reasoning| reasoning.get("summary"))
|
.and_then(|reasoning| reasoning.get("summary"))
|
||||||
.and_then(|value| value.as_str());
|
.and_then(|value| value.as_str());
|
||||||
assert_eq!(body["model"].as_str(), Some(requested_model));
|
assert_eq!(body["model"].as_str(), Some(requested_model));
|
||||||
assert_eq!(reasoning_effort, Some("future"));
|
assert_eq!(reasoning_effort, Some(expected_wire_effort));
|
||||||
assert_eq!(reasoning_summary, Some("detailed"));
|
assert_eq!(reasoning_summary, Some("detailed"));
|
||||||
|
|
||||||
Ok(())
|
Ok(())
|
||||||
|
|||||||
@@ -57,6 +57,7 @@ pub enum ReasoningEffort {
|
|||||||
XHigh,
|
XHigh,
|
||||||
Max,
|
Max,
|
||||||
Ultra,
|
Ultra,
|
||||||
|
Persistent,
|
||||||
/// A model-defined effort value that this client does not know yet.
|
/// A model-defined effort value that this client does not know yet.
|
||||||
Custom(String),
|
Custom(String),
|
||||||
}
|
}
|
||||||
@@ -73,6 +74,7 @@ impl ReasoningEffort {
|
|||||||
Self::XHigh => "xhigh",
|
Self::XHigh => "xhigh",
|
||||||
Self::Max => "max",
|
Self::Max => "max",
|
||||||
Self::Ultra => "ultra",
|
Self::Ultra => "ultra",
|
||||||
|
Self::Persistent => "persistent",
|
||||||
Self::Custom(effort) => effort,
|
Self::Custom(effort) => effort,
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
@@ -139,6 +141,7 @@ impl FromStr for ReasoningEffort {
|
|||||||
"xhigh" => Ok(Self::XHigh),
|
"xhigh" => Ok(Self::XHigh),
|
||||||
"max" => Ok(Self::Max),
|
"max" => Ok(Self::Max),
|
||||||
"ultra" => Ok(Self::Ultra),
|
"ultra" => Ok(Self::Ultra),
|
||||||
|
"persistent" => Ok(Self::Persistent),
|
||||||
"" => Err("reasoning_effort must not be empty".to_string()),
|
"" => Err("reasoning_effort must not be empty".to_string()),
|
||||||
effort => Ok(Self::Custom(effort.to_string())),
|
effort => Ok(Self::Custom(effort.to_string())),
|
||||||
}
|
}
|
||||||
@@ -1132,28 +1135,34 @@ mod tests {
|
|||||||
let serialized = to_string(&custom).expect("custom reasoning effort should serialize");
|
let serialized = to_string(&custom).expect("custom reasoning effort should serialize");
|
||||||
let serialized_max = to_string(&ReasoningEffort::Max).expect("Max should serialize");
|
let serialized_max = to_string(&ReasoningEffort::Max).expect("Max should serialize");
|
||||||
let serialized_ultra = to_string(&ReasoningEffort::Ultra).expect("Ultra should serialize");
|
let serialized_ultra = to_string(&ReasoningEffort::Ultra).expect("Ultra should serialize");
|
||||||
|
let serialized_persistent =
|
||||||
|
to_string(&ReasoningEffort::Persistent).expect("Persistent should serialize");
|
||||||
|
|
||||||
assert_eq!(
|
assert_eq!(
|
||||||
(
|
(
|
||||||
"high".parse(),
|
"high".parse(),
|
||||||
"max".parse(),
|
"max".parse(),
|
||||||
"ultra".parse(),
|
"ultra".parse(),
|
||||||
|
"persistent".parse(),
|
||||||
"future".parse(),
|
"future".parse(),
|
||||||
deserialized,
|
deserialized,
|
||||||
serialized,
|
serialized,
|
||||||
serialized_max,
|
serialized_max,
|
||||||
serialized_ultra,
|
serialized_ultra,
|
||||||
|
serialized_persistent,
|
||||||
custom.to_string(),
|
custom.to_string(),
|
||||||
),
|
),
|
||||||
(
|
(
|
||||||
Ok(ReasoningEffort::High),
|
Ok(ReasoningEffort::High),
|
||||||
Ok(ReasoningEffort::Max),
|
Ok(ReasoningEffort::Max),
|
||||||
Ok(ReasoningEffort::Ultra),
|
Ok(ReasoningEffort::Ultra),
|
||||||
|
Ok(ReasoningEffort::Persistent),
|
||||||
Ok(custom.clone()),
|
Ok(custom.clone()),
|
||||||
custom,
|
custom,
|
||||||
r#""future""#.to_string(),
|
r#""future""#.to_string(),
|
||||||
r#""max""#.to_string(),
|
r#""max""#.to_string(),
|
||||||
r#""ultra""#.to_string(),
|
r#""ultra""#.to_string(),
|
||||||
|
r#""persistent""#.to_string(),
|
||||||
"future".to_string(),
|
"future".to_string(),
|
||||||
)
|
)
|
||||||
);
|
);
|
||||||
|
|||||||
@@ -646,6 +646,7 @@ impl ChatWidget {
|
|||||||
ReasoningEffortConfig::XHigh => "Extra high".to_string(),
|
ReasoningEffortConfig::XHigh => "Extra high".to_string(),
|
||||||
ReasoningEffortConfig::Max => "Max".to_string(),
|
ReasoningEffortConfig::Max => "Max".to_string(),
|
||||||
ReasoningEffortConfig::Ultra => "Ultra".to_string(),
|
ReasoningEffortConfig::Ultra => "Ultra".to_string(),
|
||||||
|
ReasoningEffortConfig::Persistent => "Persistent".to_string(),
|
||||||
ReasoningEffortConfig::Custom(value) => value.clone(),
|
ReasoningEffortConfig::Custom(value) => value.clone(),
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|||||||
+2
-1
@@ -8,6 +8,7 @@ expression: popup
|
|||||||
2. Medium (default) Balances speed and reasoning depth for everyday tasks
|
2. Medium (default) Balances speed and reasoning depth for everyday tasks
|
||||||
› 3. High (current) Greater reasoning depth for complex problems
|
› 3. High (current) Greater reasoning depth for complex problems
|
||||||
4. Extra high Extra high reasoning depth for complex problems
|
4. Extra high Extra high reasoning depth for complex problems
|
||||||
5. More reasoning… Max and Ultra consume usage limits faster
|
5. Persistent Continue working until put to sleep
|
||||||
|
6. More reasoning… Max and Ultra consume usage limits faster
|
||||||
|
|
||||||
Press enter to confirm or esc to go back
|
Press enter to confirm or esc to go back
|
||||||
@@ -3352,6 +3352,12 @@ async fn model_reasoning_selection_popup_snapshot() {
|
|||||||
effort: ReasoningEffortConfig::Ultra,
|
effort: ReasoningEffortConfig::Ultra,
|
||||||
description: "Ultra reasoning".to_string(),
|
description: "Ultra reasoning".to_string(),
|
||||||
});
|
});
|
||||||
|
preset
|
||||||
|
.supported_reasoning_efforts
|
||||||
|
.push(ReasoningEffortPreset {
|
||||||
|
effort: ReasoningEffortConfig::Persistent,
|
||||||
|
description: "Continue working until put to sleep".to_string(),
|
||||||
|
});
|
||||||
chat.open_reasoning_popup(preset);
|
chat.open_reasoning_popup(preset);
|
||||||
|
|
||||||
let popup = render_bottom_popup(&chat, /*width*/ 80);
|
let popup = render_bottom_popup(&chat, /*width*/ 80);
|
||||||
|
|||||||
@@ -9,7 +9,8 @@ export type ModelReasoningEffort =
|
|||||||
| "high"
|
| "high"
|
||||||
| "xhigh"
|
| "xhigh"
|
||||||
| "max"
|
| "max"
|
||||||
| "ultra";
|
| "ultra"
|
||||||
|
| "persistent";
|
||||||
|
|
||||||
export type WebSearchMode = "disabled" | "cached" | "live";
|
export type WebSearchMode = "disabled" | "cached" | "live";
|
||||||
|
|
||||||
|
|||||||
@@ -229,36 +229,39 @@ describe("Codex", () => {
|
|||||||
}
|
}
|
||||||
});
|
});
|
||||||
|
|
||||||
it("passes modelReasoningEffort to exec", async () => {
|
it.each(["high", "persistent"] as const)(
|
||||||
const { url, close } = await startResponsesTestProxy({
|
"passes %s modelReasoningEffort to exec",
|
||||||
statusCode: 200,
|
async (effort) => {
|
||||||
responseBodies: [
|
const { url, close } = await startResponsesTestProxy({
|
||||||
sse(
|
statusCode: 200,
|
||||||
responseStarted("response_1"),
|
responseBodies: [
|
||||||
assistantMessage("Reasoning effort applied", "item_1"),
|
sse(
|
||||||
responseCompleted("response_1"),
|
responseStarted("response_1"),
|
||||||
),
|
assistantMessage("Reasoning effort applied", "item_1"),
|
||||||
],
|
responseCompleted("response_1"),
|
||||||
});
|
),
|
||||||
|
],
|
||||||
const { args: spawnArgs, restore } = codexExecSpy();
|
|
||||||
const { client, cleanup } = createMockClient(url);
|
|
||||||
|
|
||||||
try {
|
|
||||||
const thread = client.startThread({
|
|
||||||
modelReasoningEffort: "high",
|
|
||||||
});
|
});
|
||||||
await thread.run("apply reasoning effort");
|
|
||||||
|
|
||||||
const commandArgs = spawnArgs[0];
|
const { args: spawnArgs, restore } = codexExecSpy();
|
||||||
expect(commandArgs).toBeDefined();
|
const { client, cleanup } = createMockClient(url);
|
||||||
expectPair(commandArgs, ["--config", 'model_reasoning_effort="high"']);
|
|
||||||
} finally {
|
try {
|
||||||
cleanup();
|
const thread = client.startThread({
|
||||||
restore();
|
modelReasoningEffort: effort,
|
||||||
await close();
|
});
|
||||||
}
|
await thread.run("apply reasoning effort");
|
||||||
});
|
|
||||||
|
const commandArgs = spawnArgs[0];
|
||||||
|
expect(commandArgs).toBeDefined();
|
||||||
|
expectPair(commandArgs, ["--config", `model_reasoning_effort="${effort}"`]);
|
||||||
|
} finally {
|
||||||
|
cleanup();
|
||||||
|
restore();
|
||||||
|
await close();
|
||||||
|
}
|
||||||
|
},
|
||||||
|
);
|
||||||
|
|
||||||
it("passes networkAccessEnabled to exec", async () => {
|
it("passes networkAccessEnabled to exec", async () => {
|
||||||
const { url, close } = await startResponsesTestProxy({
|
const { url, close } = await startResponsesTestProxy({
|
||||||
|
|||||||
Reference in new issue
Block a user