service-tier

This commit is contained in:
kai hayashi
2026-02-20 12:43:01 -08:00
parent 6b1091fc92
commit b67fc0621e
23 changed files with 237 additions and 4 deletions
+20
View File
@@ -458,6 +458,9 @@
"model_reasoning_summary": {
"$ref": "#/definitions/ReasoningSummary"
},
"model_service_tier": {
"$ref": "#/definitions/ServiceTier"
},
"model_verbosity": {
"$ref": "#/definitions/Verbosity"
},
@@ -1229,6 +1232,15 @@
},
"type": "object"
},
"ServiceTier": {
"description": "Controls latency/cost class for OpenAI Responses API requests. Serialized with lowercase values to match the OpenAI API.",
"enum": [
"auto",
"default",
"flex"
],
"type": "string"
},
"ShellEnvironmentPolicyInherit": {
"oneOf": [
{
@@ -1840,6 +1852,14 @@
"model_reasoning_summary": {
"$ref": "#/definitions/ReasoningSummary"
},
"model_service_tier": {
"allOf": [
{
"$ref": "#/definitions/ServiceTier"
}
],
"description": "Optional OpenAI Responses API `service_tier` (e.g. `flex`)."
},
"model_supports_reasoning_summaries": {
"description": "Override to force-enable reasoning summaries for the configured model.",
"type": "boolean"
+6
View File
@@ -65,6 +65,7 @@ use codex_otel::OtelManager;
use codex_protocol::ThreadId;
use codex_protocol::config_types::ReasoningSummary as ReasoningSummaryConfig;
use codex_protocol::config_types::ServiceTier as ServiceTierConfig;
use codex_protocol::config_types::Verbosity as VerbosityConfig;
use codex_protocol::models::ResponseItem;
use codex_protocol::openai_models::ModelInfo;
@@ -118,6 +119,7 @@ struct ModelClientState {
provider: ModelProviderInfo,
session_source: SessionSource,
model_verbosity: Option<VerbosityConfig>,
model_service_tier: Option<ServiceTierConfig>,
enable_responses_websockets: bool,
enable_responses_websockets_v2: bool,
enable_request_compression: bool,
@@ -209,6 +211,7 @@ impl ModelClient {
provider: ModelProviderInfo,
session_source: SessionSource,
model_verbosity: Option<VerbosityConfig>,
model_service_tier: Option<ServiceTierConfig>,
enable_responses_websockets: bool,
enable_responses_websockets_v2: bool,
enable_request_compression: bool,
@@ -224,6 +227,7 @@ impl ModelClient {
provider,
session_source,
model_verbosity,
model_service_tier,
enable_responses_websockets,
enable_responses_websockets_v2,
enable_request_compression,
@@ -547,6 +551,7 @@ impl ModelClientSession {
include,
prompt_cache_key,
text,
service_tier: self.client.state.model_service_tier,
};
Ok(request)
}
@@ -1224,6 +1229,7 @@ mod tests {
provider,
session_source,
None,
None,
false,
false,
false,
+29
View File
@@ -240,6 +240,7 @@ mod tests {
use codex_api::common::OpenAiVerbosity;
use codex_api::common::TextControls;
use codex_api::create_text_param_for_request;
use codex_protocol::config_types::ServiceTier;
use pretty_assertions::assert_eq;
use super::*;
@@ -264,6 +265,7 @@ mod tests {
verbosity: Some(OpenAiVerbosity::Low),
format: None,
}),
service_tier: None,
};
let v = serde_json::to_value(&req).expect("json");
@@ -302,6 +304,7 @@ mod tests {
include: vec![],
prompt_cache_key: None,
text: Some(text_controls),
service_tier: None,
};
let v = serde_json::to_value(&req).expect("json");
@@ -338,9 +341,35 @@ mod tests {
include: vec![],
prompt_cache_key: None,
text: None,
service_tier: None,
};
let v = serde_json::to_value(&req).expect("json");
assert!(v.get("text").is_none());
}
#[test]
fn serializes_service_tier_when_set() {
let req = ResponsesApiRequest {
model: "gpt-5.1".to_string(),
instructions: "i".to_string(),
input: vec![],
tools: vec![],
tool_choice: "auto".to_string(),
parallel_tool_calls: true,
reasoning: None,
store: false,
stream: true,
include: vec![],
prompt_cache_key: None,
text: None,
service_tier: Some(ServiceTier::Flex),
};
let v = serde_json::to_value(&req).expect("json");
assert_eq!(
v.get("service_tier").and_then(|tier| tier.as_str()),
Some("flex")
);
}
}
+3
View File
@@ -1327,6 +1327,7 @@ impl Session {
session_configuration.provider.clone(),
session_configuration.session_source.clone(),
config.model_verbosity,
config.model_service_tier,
config.features.enabled(Feature::ResponsesWebsockets)
|| config.features.enabled(Feature::ResponsesWebsocketsV2),
config.features.enabled(Feature::ResponsesWebsocketsV2),
@@ -7434,6 +7435,7 @@ mod tests {
session_configuration.provider.clone(),
session_configuration.session_source.clone(),
config.model_verbosity,
config.model_service_tier,
model_info.prefer_websockets
|| config.features.enabled(Feature::ResponsesWebsockets)
|| config.features.enabled(Feature::ResponsesWebsocketsV2),
@@ -7590,6 +7592,7 @@ mod tests {
session_configuration.provider.clone(),
session_configuration.session_source.clone(),
config.model_verbosity,
config.model_service_tier,
model_info.prefer_websockets
|| config.features.enabled(Feature::ResponsesWebsockets)
|| config.features.enabled(Feature::ResponsesWebsocketsV2),
+13
View File
@@ -62,6 +62,7 @@ use codex_protocol::config_types::ForcedLoginMethod;
use codex_protocol::config_types::Personality;
use codex_protocol::config_types::ReasoningSummary;
use codex_protocol::config_types::SandboxMode;
use codex_protocol::config_types::ServiceTier;
use codex_protocol::config_types::TrustLevel;
use codex_protocol::config_types::Verbosity;
use codex_protocol::config_types::WebSearchMode;
@@ -391,6 +392,9 @@ pub struct Config {
/// Optional verbosity control for GPT-5 models (Responses API `text.verbosity`).
pub model_verbosity: Option<Verbosity>,
/// Optional OpenAI Responses API `service_tier` (e.g. `flex`).
pub model_service_tier: Option<ServiceTier>,
/// Base URL for requests to ChatGPT (as opposed to the OpenAI API).
pub chatgpt_base_url: String,
@@ -1105,6 +1109,8 @@ pub struct ConfigToml {
pub model_reasoning_summary: Option<ReasoningSummary>,
/// Optional verbosity control for GPT-5 models (Responses API `text.verbosity`).
pub model_verbosity: Option<Verbosity>,
/// Optional OpenAI Responses API `service_tier` (e.g. `flex`).
pub model_service_tier: Option<ServiceTier>,
/// Override to force-enable reasoning summaries for the configured model.
pub model_supports_reasoning_summaries: Option<bool>,
@@ -1219,6 +1225,7 @@ impl From<ConfigToml> for UserSavedConfig {
model_reasoning_effort: config_toml.model_reasoning_effort,
model_reasoning_summary: config_toml.model_reasoning_summary,
model_verbosity: config_toml.model_verbosity,
model_service_tier: config_toml.model_service_tier,
tools: config_toml.tools.map(From::from),
profile: config_toml.profile,
profiles,
@@ -2039,6 +2046,7 @@ impl Config {
model_supports_reasoning_summaries: cfg.model_supports_reasoning_summaries,
model_catalog,
model_verbosity: config_profile.model_verbosity.or(cfg.model_verbosity),
model_service_tier: config_profile.model_service_tier.or(cfg.model_service_tier),
chatgpt_base_url: config_profile
.chatgpt_base_url
.or(cfg.chatgpt_base_url)
@@ -4451,6 +4459,7 @@ approval_policy = "on-failure"
model_reasoning_effort = "high"
model_reasoning_summary = "detailed"
model_verbosity = "high"
model_service_tier = "flex"
"#;
let cfg: ConfigToml = toml::from_str(toml).expect("TOML deserialization should succeed");
@@ -4581,6 +4590,7 @@ model_verbosity = "high"
model_supports_reasoning_summaries: None,
model_catalog: None,
model_verbosity: None,
model_service_tier: None,
personality: Some(Personality::Pragmatic),
chatgpt_base_url: "https://chatgpt.com/backend-api/".to_string(),
base_instructions: None,
@@ -4700,6 +4710,7 @@ model_verbosity = "high"
model_supports_reasoning_summaries: None,
model_catalog: None,
model_verbosity: None,
model_service_tier: None,
personality: Some(Personality::Pragmatic),
chatgpt_base_url: "https://chatgpt.com/backend-api/".to_string(),
base_instructions: None,
@@ -4817,6 +4828,7 @@ model_verbosity = "high"
model_supports_reasoning_summaries: None,
model_catalog: None,
model_verbosity: None,
model_service_tier: None,
personality: Some(Personality::Pragmatic),
chatgpt_base_url: "https://chatgpt.com/backend-api/".to_string(),
base_instructions: None,
@@ -4920,6 +4932,7 @@ model_verbosity = "high"
model_supports_reasoning_summaries: None,
model_catalog: None,
model_verbosity: Some(Verbosity::High),
model_service_tier: Some(ServiceTier::Flex),
personality: Some(Personality::Pragmatic),
chatgpt_base_url: "https://chatgpt.com/backend-api/".to_string(),
base_instructions: None,
+3
View File
@@ -8,6 +8,7 @@ use crate::config::types::WindowsToml;
use crate::protocol::AskForApproval;
use codex_protocol::config_types::ReasoningSummary;
use codex_protocol::config_types::SandboxMode;
use codex_protocol::config_types::ServiceTier;
use codex_protocol::config_types::Verbosity;
use codex_protocol::config_types::WebSearchMode;
use codex_protocol::openai_models::ReasoningEffort;
@@ -26,6 +27,7 @@ pub struct ConfigProfile {
pub model_reasoning_effort: Option<ReasoningEffort>,
pub model_reasoning_summary: Option<ReasoningSummary>,
pub model_verbosity: Option<Verbosity>,
pub model_service_tier: Option<ServiceTier>,
pub personality: Option<Personality>,
pub chatgpt_base_url: Option<String>,
/// Optional path to a file containing model instructions.
@@ -65,6 +67,7 @@ impl From<ConfigProfile> for codex_app_server_protocol::Profile {
model_reasoning_effort: config_profile.model_reasoning_effort,
model_reasoning_summary: config_profile.model_reasoning_summary,
model_verbosity: config_profile.model_verbosity,
model_service_tier: config_profile.model_service_tier,
chatgpt_base_url: config_profile.chatgpt_base_url,
}
}
+3
View File
@@ -91,6 +91,7 @@ async fn responses_stream_includes_subagent_header_on_review() {
provider.clone(),
session_source,
config.model_verbosity,
config.model_service_tier,
false,
false,
false,
@@ -197,6 +198,7 @@ async fn responses_stream_includes_subagent_header_on_other() {
provider.clone(),
session_source,
config.model_verbosity,
config.model_service_tier,
false,
false,
false,
@@ -302,6 +304,7 @@ async fn responses_respects_model_info_overrides_from_config() {
provider.clone(),
session_source,
config.model_verbosity,
config.model_service_tier,
false,
false,
false,
+1
View File
@@ -1353,6 +1353,7 @@ async fn azure_responses_request_includes_store_and_reasoning_ids() {
provider.clone(),
SessionSource::Exec,
config.model_verbosity,
config.model_service_tier,
false,
false,
false,
@@ -1251,6 +1251,7 @@ async fn websocket_harness_with_options(
provider.clone(),
SessionSource::Exec,
config.model_verbosity,
config.model_service_tier,
websocket_enabled,
websocket_v2_enabled,
false,