Enable concurrent reasoning summaries by default (#62903)

## What changed

Mark `concurrent_reasoning_summaries` as stable and enable it by default.
When requesting reasoning summaries from OpenAI, Codex now sets
`stream_options.reasoning_summary_delivery` to `sequential_cutoff` by
default.

## Testing

Exercise summary rendering with the default configuration and update
request assertions and snapshots to expect `sequential_cutoff`. Explicitly
disable the feature in tests that exercise legacy reasoning events.

GitOrigin-RevId: 04310fd271ced11c3a22e8ff6275a387c9c22fc2
This commit is contained in:
Eric Traut
2026-11-21 05:42:40 +00:00
committed by copyberry
parent db72403243
commit 58952e0c39
15 changed files with 33 additions and 20 deletions
@@ -14,6 +14,7 @@ Settings:
service_tier: "priority"
store: false
stream: true
stream_options: {"reasoning_summary_delivery":"sequential_cutoff"}
tool_choice: "auto"
-- request 1 (turn) --
00:additional_tools/developer (3; hash=ba5ddd25650ac9fa):
-3
View File
@@ -3027,9 +3027,6 @@ async fn configured_reasoning_summary_is_sent() -> anyhow::Result<()> {
let TestCodex { codex, .. } = test_codex()
.with_config(|config| {
config.model_reasoning_summary = Some(ReasoningSummary::Concise);
let _ = config
.features
.enable(Feature::ConcurrentReasoningSummaries);
})
.build(&server)
.await?;
@@ -56,6 +56,10 @@ async fn codex_delegate_ignores_legacy_deltas(explicit_guardian: bool) {
mount_sse_sequence(&server, vec![sse_stream]).await;
let mut builder = test_codex().with_config(move |config| {
config
.features
.disable(Feature::ConcurrentReasoningSummaries)
.expect("disable concurrent summaries");
if explicit_guardian {
config.guardian_model_provider = Some(Arc::new(GuardianInferenceProvider {
id: "guardian".to_string(),
+10 -10
View File
@@ -1164,7 +1164,15 @@ async fn reasoning_content_delta_has_item_metadata() -> anyhow::Result<()> {
let server = start_mock_server().await;
let TestCodex { codex, .. } = test_codex().build(&server).await?;
let TestCodex { codex, .. } = test_codex()
.with_config(|config| {
config
.features
.disable(Feature::ConcurrentReasoningSummaries)
.expect("disable concurrent summaries");
})
.build(&server)
.await?;
let stream = sse(vec![
ev_response_created("resp-1"),
@@ -1209,15 +1217,7 @@ async fn sequential_cutoff_renders_done_summaries_for_active_reasoning_item() ->
let server = start_mock_server().await;
let TestCodex { codex, .. } = test_codex()
.with_config(|config| {
config
.features
.enable(Feature::ConcurrentReasoningSummaries)
.expect("test config should allow feature update");
})
.build(&server)
.await?;
let TestCodex { codex, .. } = test_codex().build(&server).await?;
let summary_done = |item_id: &str, summary_index: i64, text: &str| {
serde_json::json!({
"type": "response.reasoning_summary_text.done",
@@ -4571,6 +4571,10 @@ async fn conversation_relays_only_public_reasoning_as_quiet_status(enabled: bool
config.realtime.version = RealtimeWsVersion::V3;
config.model_reasoning_summary =
Some(codex_protocol::config_types::ReasoningSummary::None);
config
.features
.disable(Feature::ConcurrentReasoningSummaries)
.expect("disable concurrent summaries");
}
});
let test = builder.build_with_streaming_server(&api_server).await?;
@@ -13,6 +13,7 @@ Settings:
reasoning: {"summary":"auto"}
store: false
stream: true
stream_options: {"reasoning_summary_delivery":"sequential_cutoff"}
tool_choice: "auto"
tools (11; hash=067d271a36f2eb01):
- function/exec_command: Runs a command in a PTY, returning output or a session ID for ongoing interaction.; args=[cmd, justification, login, max_output_tokens, prefix_rule, sandbox_permissions, shell, tty, workdir, yield_time_ms]; hash=24f4b6739dd5d940
@@ -14,6 +14,7 @@ Settings:
reasoning: {"context":"all_turns","summary":"auto"}
store: false
stream: true
stream_options: {"reasoning_summary_delivery":"sequential_cutoff"}
tool_choice: "auto"
-- request 1 (turn) --
00:additional_tools/developer (2; hash=fb72c6881199d93a):
@@ -13,6 +13,7 @@ Settings:
reasoning: {"context":"all_turns","summary":"auto"}
store: false
stream: true
stream_options: {"reasoning_summary_delivery":"sequential_cutoff"}
tool_choice: "auto"
-- request 1 (request) --
00:additional_tools/developer (2; hash=ba09b4f2f05f1575):
@@ -13,6 +13,7 @@ Settings:
reasoning: {"context":"all_turns","summary":"auto"}
store: false
stream: true
stream_options: {"reasoning_summary_delivery":"sequential_cutoff"}
tool_choice: "auto"
-- request 1 (request) --
00:additional_tools/developer (2; hash=ba09b4f2f05f1575):
@@ -22,7 +22,7 @@ Scenario: Guardian reviews a host-owned orchestrator connector, then the parent
03:message/user:
List today's calendar events.
## Window 2 (after request 1: input diverged at item 00, settings changed (model, parallel_tool_calls, prompt_cache_key, reasoning, text, tool_choice, tools))
## Window 2 (after request 1: input diverged at item 00, settings changed (model, parallel_tool_calls, prompt_cache_key, reasoning, stream_options, text, tool_choice, tools))
-- request 2 (request) --
00:additional_tools/developer (0; hash=09612b07b5ecb5a5):
01:message/developer:
@@ -87,7 +87,7 @@ Scenario: Guardian reviews a host-owned orchestrator connector, then the parent
[07] >>> APPROVAL REQUEST END
## Window 3 (after request 2: input diverged at item 00, settings changed (model, parallel_tool_calls, prompt_cache_key, reasoning, text, tool_choice, tools))
## Window 3 (after request 2: input diverged at item 00, settings changed (model, parallel_tool_calls, prompt_cache_key, reasoning, stream_options, text, tool_choice, tools))
-- request 3 (request) --
00:message/developer:
<GUARDIAN_INSTRUCTIONS>
@@ -13,6 +13,7 @@ Settings:
reasoning: {"effort":"medium","summary":"auto"}
store: false
stream: true
stream_options: {"reasoning_summary_delivery":"sequential_cutoff"}
tool_choice: "auto"
tools (6; hash=6d27d4ee152c61c9):
- function/exec_command: Runs a command in a PTY, returning output or a session ID for ongoing interaction.; args=[cmd, justification, login, max_output_tokens, prefix_rule, sandbox_permissions, shell, tty, workdir, yield_time_ms]; hash=24f4b6739dd5d940
@@ -13,6 +13,7 @@ Settings:
reasoning: {"context":"all_turns","summary":"auto"}
store: false
stream: true
stream_options: {"reasoning_summary_delivery":"sequential_cutoff"}
tool_choice: "auto"
-- request 1 (turn) --
00:additional_tools/developer (2; hash=86ed27a8ce319b21):
@@ -13,6 +13,7 @@ Settings:
reasoning: {"summary":"auto"}
store: false
stream: true
stream_options: {"reasoning_summary_delivery":"sequential_cutoff"}
tool_choice: "auto"
tools (6; hash=04baa2cc544b973a):
- function/exec_command: Runs a command in a PTY, returning output or a session ID for ongoing interaction.; args=[cmd, justification, login, max_output_tokens, prefix_rule, sandbox_permissions, shell, tty, workdir, yield_time_ms]; hash=24f4b6739dd5d940
+2 -2
View File
@@ -2148,8 +2148,8 @@ pub const FEATURES: &[FeatureSpec] = &[
FeatureSpec {
id: Feature::ConcurrentReasoningSummaries,
key: "concurrent_reasoning_summaries",
stage: Stage::UnderDevelopment,
default_enabled: false,
stage: Stage::Stable,
default_enabled: true,
},
FeatureSpec {
id: Feature::OutputTokenReplay,
@@ -14,7 +14,7 @@ async fn reasoning_defaults_reach_responses() -> Result<()> {
(
"",
json!("concise"),
json!(null),
json!({"reasoning_summary_delivery": "sequential_cutoff"}),
ThreadParamsMode::Embedded,
),
(
@@ -32,7 +32,7 @@ async fn reasoning_defaults_reach_responses() -> Result<()> {
(
"model_reasoning_summary = 'detailed'",
json!("detailed"),
json!(null),
json!({"reasoning_summary_delivery": "sequential_cutoff"}),
ThreadParamsMode::Embedded,
),
(
@@ -50,7 +50,7 @@ async fn reasoning_defaults_reach_responses() -> Result<()> {
(
"model_reasoning_summary = 'detailed'\nmodel_verbosity = 'high'\nweb_search = 'live'",
json!("detailed"),
json!(null),
json!({"reasoning_summary_delivery": "sequential_cutoff"}),
ThreadParamsMode::Remote,
),
] {