feat(llm, openai): Add GPT-5.6, max reasoning, prompt caching

.jp/config.toml

@@ -33,10 +33,14 @@ opus6 = "anthropic/claude-opus-4-6"
 haiku = "anthropic/claude-haiku-4-5"
 zai = "cerebras/zai-glm-4.7"
 # OpenAI aliases
-openai = "openai/gpt-5.5"
-chatgpt = "openai/gpt-5.5"
-gpt = "openai/gpt-5.5"
-gpt5 = "openai/gpt-5.5"
+openai = "openai/gpt-5.6-sol"
+chatgpt = "openai/gpt-5.6-sol"
+gpt = "openai/gpt-5.6-sol"
+gpt5 = "openai/gpt-5.6-sol"
+gpt55 = "openai/gpt-5.5"
+sol = "openai/gpt-5.6-sol"
+terra = "openai/gpt-5.6-terra"
+luna = "openai/gpt-5.6-luna"
 gpt5-pro = "openai/gpt-5.5-pro"
 gpt-pro = "openai/gpt-5.5-pro"
 gpt5-mini = "openai/gpt-5.4-mini"

crates/jp_llm/src/model.rs

@@ -147,12 +147,13 @@ impl ModelDetails {
                 medium,
                 high,
                 xhigh,
+                max,
             }) => match config {
                 // Off, so disabled.
                 Some(ReasoningConfig::Off) => None,
 
                 // Auto configured, so use medium effort if the model supports
-                // it, otherwise high or low.
+                // it, otherwise the nearest supported level.
                 None | Some(ReasoningConfig::Auto) => Some(CustomReasoningConfig {
                     effort: if medium {
                         ReasoningEffort::Medium
@@ -162,6 +163,8 @@ impl ModelDetails {
                         ReasoningEffort::XHigh
                     } else if low {
                         ReasoningEffort::Low
+                    } else if max {
+                        ReasoningEffort::Max
                     } else {
                         ReasoningEffort::Xlow
                     },
@@ -262,6 +265,10 @@ pub enum ReasoningDetails {
 
         /// Whether the model supports extremely high effort reasoning.
         xhigh: bool,
+
+        /// Whether the model supports maximum effort reasoning, with no
+        /// constraints on token spending.
+        max: bool,
     },
 
     /// Adaptive reasoning support.
@@ -298,6 +305,7 @@ impl ReasoningDetails {
         medium: bool,
         high: bool,
         xhigh: bool,
+        max: bool,
     ) -> Self {
         Self::Leveled {
             none,
@@ -306,6 +314,7 @@ impl ReasoningDetails {
             medium,
             high,
             xhigh,
+            max,
         }
     }
 
@@ -352,6 +361,7 @@ impl ReasoningDetails {
                 medium,
                 high,
                 xhigh,
+                max,
             } => {
                 if *none {
                     Some(ReasoningEffort::None)
@@ -365,6 +375,8 @@ impl ReasoningDetails {
                     Some(ReasoningEffort::High)
                 } else if *xhigh {
                     Some(ReasoningEffort::XHigh)
+                } else if *max {
+                    Some(ReasoningEffort::Max)
                 } else {
                     None
                 }
@@ -373,6 +385,15 @@ impl ReasoningDetails {
         }
     }
 
+    /// Returns `true` if the model supports the `max` reasoning effort level.
+    #[must_use]
+    pub fn supports_max_effort(&self) -> bool {
+        matches!(
+            self,
+            Self::Leveled { max: true, .. } | Self::Adaptive { max: true, .. }
+        )
+    }
+
     #[must_use]
     pub fn is_unsupported(&self) -> bool {
         matches!(self, Self::Unsupported)
@@ -393,3 +414,7 @@ impl ReasoningDetails {
         matches!(self, Self::Adaptive { .. })
     }
 }
+
+#[cfg(test)]
+#[path = "model_tests.rs"]
+mod tests;

crates/jp_llm/src/model_tests.rs

@@ -0,0 +1,43 @@
+use jp_config::model::parameters::{ReasoningConfig, ReasoningEffort};
+
+use super::{ModelDetails, ReasoningDetails};
+
+mod custom_reasoning_config {
+    use super::*;
+
+    fn model(reasoning: ReasoningDetails) -> ModelDetails {
+        let mut details = ModelDetails::empty("openai/test-model".parse().unwrap());
+        details.reasoning = Some(reasoning);
+        details
+    }
+
+    /// A leveled model whose only supported level is `max` resolves `Auto` to
+    /// `max` instead of falling through to an unsupported level.
+    #[test]
+    fn auto_on_max_only_model_selects_max() {
+        let details = model(ReasoningDetails::leveled(
+            false, false, false, false, false, false, true,
+        ));
+
+        let config = details
+            .custom_reasoning_config(Some(ReasoningConfig::Auto))
+            .unwrap();
+
+        assert_eq!(config.effort, ReasoningEffort::Max);
+    }
+
+    /// `max` is a last resort: any lower supported level wins in the `Auto`
+    /// selection.
+    #[test]
+    fn auto_prefers_lower_levels_over_max() {
+        let details = model(ReasoningDetails::leveled(
+            false, false, true, false, false, false, true,
+        ));
+
+        let config = details
+            .custom_reasoning_config(Some(ReasoningConfig::Auto))
+            .unwrap();
+
+        assert_eq!(config.effort, ReasoningEffort::Low);
+    }
+}

crates/jp_llm/src/provider/cerebras.rs

@@ -228,7 +228,7 @@ fn map_model(id: &str) -> Result<ModelDetails> {
             context_window: Some(131_072),
             max_output_tokens: Some(40_960),
             reasoning: Some(ReasoningDetails::leveled(
-                false, false, true, true, true, false,
+                false, false, true, true, true, false, false,
             )),
             knowledge_cutoff: None,
             deprecated: None,
@@ -242,7 +242,7 @@ fn map_model(id: &str) -> Result<ModelDetails> {
             max_output_tokens: Some(40_960),
             // Reasoning is enabled by default; only `none` disables it.
             reasoning: Some(ReasoningDetails::leveled(
-                true, false, false, false, false, false,
+                true, false, false, false, false, false, false,
             )),
             knowledge_cutoff: None,
             deprecated: None,

crates/jp_llm/src/provider/google.rs

@@ -245,6 +245,7 @@ fn create_request(
                         medium,
                         high,
                         xhigh: _,
+                        max: _,
                     } => {
                         let level = config
                             .effort
@@ -408,7 +409,7 @@ fn map_model(model: types::Model) -> ModelDetails {
                 context_window,
                 max_output_tokens,
                 reasoning: Some(ReasoningDetails::leveled(
-                    false, false, true, true, true, false,
+                    false, false, true, true, true, false, false,
                 )),
                 knowledge_cutoff: Some(NaiveDate::from_ymd_opt(2025, 1, 1).unwrap()),
                 deprecated: Some(ModelDeprecation::Active),
@@ -422,7 +423,7 @@ fn map_model(model: types::Model) -> ModelDetails {
             context_window,
             max_output_tokens,
             reasoning: Some(ReasoningDetails::leveled(
-                false, false, true, false, true, false,
+                false, false, true, false, true, false, false,
             )),
             knowledge_cutoff: Some(NaiveDate::from_ymd_opt(2025, 1, 1).unwrap()),
             deprecated: Some(ModelDeprecation::Active),
@@ -435,7 +436,7 @@ fn map_model(model: types::Model) -> ModelDetails {
             context_window,
             max_output_tokens,
             reasoning: Some(ReasoningDetails::leveled(
-                false, true, true, true, true, false,
+                false, true, true, true, true, false, false,
             )),
             knowledge_cutoff: Some(NaiveDate::from_ymd_opt(2025, 1, 1).unwrap()),
             deprecated: Some(ModelDeprecation::Active),