agent/codex
2
0
Fork 0
mirror of https://github.com/pchuan98/codex.git synced 2026-07-01 00:31:56 +08:00
Code Issues Packages Projects Releases Wiki Activity

[codex] Support model-defined reasoning efforts (#26444)

Browse Source 
## Summary
- accept non-empty model-defined reasoning effort values while
preserving built-in effort behavior
- propagate the non-Copy effort type through core, app-server, TUI,
telemetry, and persistence call sites
- preserve string wire encoding and expose an open-string schema for
clients
- update model selection and shortcut behavior for model-advertised
effort values

## Root cause
`ReasoningEffort` gained a string-backed custom variant, so it could no
longer implement `Copy` or rely on derived closed-enum serialization.
Existing consumers still moved effort values from shared references and
assumed a fixed built-in value set.

## Validation
- `just fmt`
- Local tests and compilation were not run per request; relying on CI.
This commit is contained in:
Ahmed Ibrahim
2026-06-04 13:36:24 -07:00
committed by GitHub
parent 4a70e0ac1b
commit 8ac304c299
97 changed files with 685 additions and 581 deletions
Download Patch File Download Diff File Expand all files Collapse all files
codex-rs/core/tests/responses_headers.rs
+3 -3
View File
@@ -74,7 +74,7 @@ async fn responses_stream_includes_subagent_header_on_review() {
let mut config = load_default_config_for_test(&codex_home).await;
config.model_provider_id = provider.name.clone();
config.model_provider = provider.clone();
let effort = config.model_reasoning_effort;
let effort = config.model_reasoning_effort.clone();
let summary = config.model_reasoning_summary;
let model = codex_core::test_support::get_model_offline(config.model.as_deref());
config.model = Some(model.clone());
@@ -202,7 +202,7 @@ async fn responses_stream_includes_subagent_header_on_other() {
let mut config = load_default_config_for_test(&codex_home).await;
config.model_provider_id = provider.name.clone();
config.model_provider = provider.clone();
let effort = config.model_reasoning_effort;
let effort = config.model_reasoning_effort.clone();
let summary = config.model_reasoning_summary;
let model = codex_core::test_support::get_model_offline(config.model.as_deref());
config.model = Some(model.clone());
@@ -318,7 +318,7 @@ async fn responses_respects_model_info_overrides_from_config() {
config.model_provider = provider.clone();
config.model_supports_reasoning_summaries = Some(true);
config.model_reasoning_summary = Some(ReasoningSummary::Detailed);
let effort = config.model_reasoning_effort;
let effort = config.model_reasoning_effort.clone();
let summary = config.model_reasoning_summary;
let model = config.model.clone().expect("model configured");
let config = Arc::new(config);
codex-rs/core/tests/suite/client.rs
+2 -2
View File
@@ -876,7 +876,7 @@ async fn send_provider_auth_request(server: &MockServer, auth: ModelProviderAuth
let mut config = load_default_config_for_test(&codex_home).await;
config.model_provider_id = provider.name.clone();
config.model_provider = provider.clone();
let effort = config.model_reasoning_effort;
let effort = config.model_reasoning_effort.clone();
let summary = config.model_reasoning_summary;
let model = codex_core::test_support::get_model_offline(config.model.as_deref());
config.model = Some(model.clone());
@@ -2332,7 +2332,7 @@ async fn azure_responses_request_includes_store_and_reasoning_ids() {
let mut config = load_default_config_for_test(&codex_home).await;
config.model_provider_id = provider.name.clone();
config.model_provider = provider.clone();
let effort = config.model_reasoning_effort;
let effort = config.model_reasoning_effort.clone();
let summary = config.model_reasoning_summary;
let model = codex_core::test_support::get_model_offline(config.model.as_deref());
config.model = Some(model.clone());
codex-rs/core/tests/suite/client_websockets.rs
+12 -12
View File
@@ -343,7 +343,7 @@ async fn responses_websocket_request_prewarm_reuses_connection() {
&prompt,
&harness.model_info,
&harness.session_telemetry,
harness.effort,
harness.effort.clone(),
harness.summary,
/*service_tier*/ None,
/*turn_metadata_header*/ None,
@@ -397,7 +397,7 @@ async fn responses_websocket_request_prewarm_traces_logical_request() {
&prompt,
&harness.model_info,
&harness.session_telemetry,
harness.effort,
harness.effort.clone(),
harness.summary,
/*service_tier*/ None,
/*turn_metadata_header*/ None,
@@ -442,7 +442,7 @@ async fn responses_websocket_request_prewarm_traces_logical_request() {
&prompt,
&harness.model_info,
&harness.session_telemetry,
harness.effort,
harness.effort.clone(),
harness.summary,
/*service_tier*/ None,
/*turn_metadata_header*/ None,
@@ -534,7 +534,7 @@ async fn responses_websocket_preconnect_is_reused_even_with_header_changes() {
&prompt,
&harness.model_info,
&harness.session_telemetry,
harness.effort,
harness.effort.clone(),
harness.summary,
/*service_tier*/ None,
/*turn_metadata_header*/ None,
@@ -573,7 +573,7 @@ async fn responses_websocket_request_prewarm_is_reused_even_with_header_changes(
&prompt,
&harness.model_info,
&harness.session_telemetry,
harness.effort,
harness.effort.clone(),
harness.summary,
/*service_tier*/ None,
/*turn_metadata_header*/ None,
@@ -585,7 +585,7 @@ async fn responses_websocket_request_prewarm_is_reused_even_with_header_changes(
&prompt,
&harness.model_info,
&harness.session_telemetry,
harness.effort,
harness.effort.clone(),
harness.summary,
/*service_tier*/ None,
/*turn_metadata_header*/ None,
@@ -639,7 +639,7 @@ async fn responses_websocket_prewarm_uses_v2_when_provider_supports_websockets()
&prompt,
&harness.model_info,
&harness.session_telemetry,
harness.effort,
harness.effort.clone(),
harness.summary,
/*service_tier*/ None,
/*turn_metadata_header*/ None,
@@ -988,7 +988,7 @@ async fn responses_websocket_emits_reasoning_included_event() {
&prompt,
&harness.model_info,
&harness.session_telemetry,
harness.effort,
harness.effort.clone(),
harness.summary,
/*service_tier*/ None,
/*turn_metadata_header*/ None,
@@ -1062,7 +1062,7 @@ async fn responses_websocket_emits_rate_limit_events() {
&prompt,
&harness.model_info,
&harness.session_telemetry,
harness.effort,
harness.effort.clone(),
harness.summary,
/*service_tier*/ None,
/*turn_metadata_header*/ None,
@@ -1718,7 +1718,7 @@ async fn responses_websocket_v2_after_error_uses_full_create_without_previous_re
&prompt_two,
&harness.model_info,
&harness.session_telemetry,
harness.effort,
harness.effort.clone(),
harness.summary,
/*service_tier*/ None,
/*turn_metadata_header*/ None,
@@ -1806,7 +1806,7 @@ async fn responses_websocket_v2_surfaces_terminal_error_without_close_handshake(
&prompt_two,
&harness.model_info,
&harness.session_telemetry,
harness.effort,
harness.effort.clone(),
harness.summary,
/*service_tier*/ None,
/*turn_metadata_header*/ None,
@@ -2073,7 +2073,7 @@ async fn stream_until_complete_with_request_metadata(
prompt,
&harness.model_info,
&harness.session_telemetry,
harness.effort,
harness.effort.clone(),
harness.summary,
service_tier.map(|service_tier| service_tier.request_value().to_string()),
turn_metadata_header,
codex-rs/core/tests/suite/model_switching.rs
+1 -1
View File
@@ -58,7 +58,7 @@ fn read_only_user_turn(test: &TestCodex, items: Vec<UserInput>, model: String) -
mode: codex_protocol::config_types::ModeKind::Default,
settings: codex_protocol::config_types::Settings {
model,
reasoning_effort: test.config.model_reasoning_effort,
reasoning_effort: test.config.model_reasoning_effort.clone(),
developer_instructions: None,
},
}),
codex-rs/core/tests/suite/model_visible_layout.rs
+5 -5
View File
@@ -135,7 +135,7 @@ async fn snapshot_model_visible_layout_turn_overrides() -> Result<()> {
mode: codex_protocol::config_types::ModeKind::Default,
settings: codex_protocol::config_types::Settings {
model: test.session_configured.model.clone(),
reasoning_effort: test.config.model_reasoning_effort,
reasoning_effort: test.config.model_reasoning_effort.clone(),
developer_instructions: None,
},
}),
@@ -172,7 +172,7 @@ async fn snapshot_model_visible_layout_turn_overrides() -> Result<()> {
mode: codex_protocol::config_types::ModeKind::Default,
settings: codex_protocol::config_types::Settings {
model: test.session_configured.model.clone(),
reasoning_effort: test.config.model_reasoning_effort,
reasoning_effort: test.config.model_reasoning_effort.clone(),
developer_instructions: None,
},
}),
@@ -261,7 +261,7 @@ async fn snapshot_model_visible_layout_cwd_change_does_not_refresh_agents() -> R
mode: codex_protocol::config_types::ModeKind::Default,
settings: codex_protocol::config_types::Settings {
model: test.session_configured.model.clone(),
reasoning_effort: test.config.model_reasoning_effort,
reasoning_effort: test.config.model_reasoning_effort.clone(),
developer_instructions: None,
},
}),
@@ -295,7 +295,7 @@ async fn snapshot_model_visible_layout_cwd_change_does_not_refresh_agents() -> R
mode: codex_protocol::config_types::ModeKind::Default,
settings: codex_protocol::config_types::Settings {
model: test.session_configured.model.clone(),
reasoning_effort: test.config.model_reasoning_effort,
reasoning_effort: test.config.model_reasoning_effort.clone(),
developer_instructions: None,
},
}),
@@ -422,7 +422,7 @@ async fn snapshot_model_visible_layout_resume_with_personality_change() -> Resul
mode: codex_protocol::config_types::ModeKind::Default,
settings: codex_protocol::config_types::Settings {
model: resumed.session_configured.model.clone(),
reasoning_effort: resumed.config.model_reasoning_effort,
reasoning_effort: resumed.config.model_reasoning_effort.clone(),
developer_instructions: None,
},
}),
codex-rs/core/tests/suite/personality.rs
+1 -1
View File
@@ -79,7 +79,7 @@ fn read_only_text_turn_with_personality(
mode: codex_protocol::config_types::ModeKind::Default,
settings: codex_protocol::config_types::Settings {
model,
reasoning_effort: test.config.model_reasoning_effort,
reasoning_effort: test.config.model_reasoning_effort.clone(),
developer_instructions: None,
},
}),
codex-rs/core/tests/suite/prompt_caching.rs
+3 -3
View File
@@ -870,7 +870,7 @@ async fn send_user_turn_with_no_changes_does_not_send_environment_context() -> a
let default_approval_policy = config.permissions.approval_policy.value();
let default_sandbox_policy = &config.legacy_sandbox_policy();
let default_model = session_configured.model;
let default_effort = config.model_reasoning_effort;
let default_effort = config.model_reasoning_effort.clone();
let default_summary = config.model_reasoning_summary;
codex
@@ -892,7 +892,7 @@ async fn send_user_turn_with_no_changes_does_not_send_environment_context() -> a
mode: codex_protocol::config_types::ModeKind::Default,
settings: codex_protocol::config_types::Settings {
model: default_model.clone(),
reasoning_effort: default_effort,
reasoning_effort: default_effort.clone(),
developer_instructions: None,
},
}),
@@ -1013,7 +1013,7 @@ async fn send_user_turn_with_changes_sends_environment_context() -> anyhow::Resu
let default_approval_policy = config.permissions.approval_policy.value();
let default_sandbox_policy = &config.legacy_sandbox_policy();
let default_model = session_configured.model;
let default_effort = config.model_reasoning_effort;
let default_effort = config.model_reasoning_effort.clone();
let default_summary = config.model_reasoning_summary;
codex
codex-rs/core/tests/suite/remote_models.rs
+6 -5
View File
@@ -318,7 +318,7 @@ async fn remote_models_use_context_window_when_config_override_is_absent() -> Re
}
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn remote_models_long_model_slug_is_sent_with_high_reasoning() -> Result<()> {
async fn remote_models_long_model_slug_is_sent_with_custom_reasoning() -> Result<()> {
skip_if_no_network!(Ok(()));
skip_if_sandbox!(Ok(()));
@@ -331,15 +331,16 @@ async fn remote_models_long_model_slug_is_sent_with_high_reasoning() -> Result<(
/*priority*/ 1_000,
TruncationPolicyConfig::bytes(/*limit*/ 10_000),
);
remote_model.default_reasoning_level = Some(ReasoningEffort::High);
let custom_reasoning_effort = ReasoningEffort::Custom("max".to_string());
remote_model.default_reasoning_level = Some(custom_reasoning_effort.clone());
remote_model.supported_reasoning_levels = vec![
ReasoningEffortPreset {
effort: ReasoningEffort::Medium,
description: ReasoningEffort::Medium.to_string(),
},
ReasoningEffortPreset {
effort: ReasoningEffort::High,
description: ReasoningEffort::High.to_string(),
effort: custom_reasoning_effort.clone(),
description: custom_reasoning_effort.to_string(),
},
];
remote_model.supports_reasoning_summaries = true;
@@ -393,7 +394,7 @@ async fn remote_models_long_model_slug_is_sent_with_high_reasoning() -> Result<(
.and_then(|reasoning| reasoning.get("summary"))
.and_then(|value| value.as_str());
assert_eq!(body["model"].as_str(), Some(requested_model));
assert_eq!(reasoning_effort, Some("high"));
assert_eq!(reasoning_effort, Some("max"));
assert_eq!(reasoning_summary, Some("detailed"));
Ok(())
codex-rs/core/tests/suite/resume_warning.rs
+1 -1
View File
@@ -41,7 +41,7 @@ fn resume_history(
collaboration_mode: None,
multi_agent_version: None,
realtime_active: None,
effort: config.model_reasoning_effort,
effort: config.model_reasoning_effort.clone(),
summary: config
.model_reasoning_summary
.unwrap_or(ReasoningSummary::Auto),
codex-rs/core/tests/suite/safety_check_downgrade.rs
+1 -1
View File
@@ -55,7 +55,7 @@ fn disabled_text_turn(test: &TestCodex, text: &str) -> Op {
mode: codex_protocol::config_types::ModeKind::Default,
settings: codex_protocol::config_types::Settings {
model: REQUESTED_MODEL.to_string(),
reasoning_effort: test.config.model_reasoning_effort,
reasoning_effort: test.config.model_reasoning_effort.clone(),
developer_instructions: None,
},
}),