diff --git a/relay-event-normalization/src/normalize/mod.rs b/relay-event-normalization/src/normalize/mod.rs
index a8637234b28..bfd49c5e1e9 100644
--- a/relay-event-normalization/src/normalize/mod.rs
+++ b/relay-event-normalization/src/normalize/mod.rs
@@ -261,27 +261,24 @@ impl ModelCosts {
     }
 
     /// Gets the cost per token, if defined for the given model.
-    pub fn cost_per_token(&self, model_id: &str) -> Option<ModelCostV2> {
+    pub fn cost_per_token(&self, model_id: &str) -> Option<&ModelCostV2> {
         if !self.is_enabled() {
             return None;
         }
 
         // First try exact match by creating a Pattern from the model_id
-        let exact_key = Pattern::new(model_id).ok()?;
-        if self.models.contains_key(&exact_key) {
-            return self.models.get(&exact_key).copied();
+        if let Some(value) = self.models.get(model_id) {
+            return Some(value);
         }
 
         // if there is not a direct match, try to find the match using a pattern
-        let cost = self.models.iter().find_map(|(key, value)| {
+        self.models.iter().find_map(|(key, value)| {
             if key.is_match(model_id) {
                 Some(value)
             } else {
                 None
             }
-        });
-
-        cost.copied()
+        })
     }
 }
 
@@ -481,7 +478,7 @@ mod tests {
         let cost = v2_config.cost_per_token("gpt-4").unwrap();
         assert_eq!(
             cost,
-            ModelCostV2 {
+            &ModelCostV2 {
                 input_per_token: 0.03,
                 output_per_token: 0.06,
                 output_reasoning_per_token: 0.12,
@@ -523,7 +520,7 @@ mod tests {
         let cost = v2_config.cost_per_token("gpt-4-v1").unwrap();
         assert_eq!(
             cost,
-            ModelCostV2 {
+            &ModelCostV2 {
                 input_per_token: 0.03,
                 output_per_token: 0.06,
                 output_reasoning_per_token: 0.12,
@@ -534,7 +531,7 @@ mod tests {
         let cost = v2_config.cost_per_token("gpt-4-2xxx").unwrap();
         assert_eq!(
             cost,
-            ModelCostV2 {
+            &ModelCostV2 {
                 input_per_token: 0.0007,
                 output_per_token: 0.0008,
                 output_reasoning_per_token: 0.0016,
diff --git a/relay-event-normalization/src/normalize/span/ai.rs b/relay-event-normalization/src/normalize/span/ai.rs
index 7041f7b2bba..fb4148cec2a 100644
--- a/relay-event-normalization/src/normalize/span/ai.rs
+++ b/relay-event-normalization/src/normalize/span/ai.rs
@@ -7,7 +7,7 @@ use relay_protocol::{Annotated, Getter, Value};
 
 /// Calculates the cost of an AI model based on the model cost and the tokens used.
 /// Calculated cost is in US dollars.
-fn extract_ai_model_cost_data(model_cost: Option<ModelCostV2>, data: &mut SpanData) {
+fn extract_ai_model_cost_data(model_cost: Option<&ModelCostV2>, data: &mut SpanData) {
     let cost_per_token = match model_cost {
         Some(v) => v,
         None => return,
@@ -123,9 +123,7 @@ fn extract_ai_data(span: &mut Span, ai_model_costs: &ModelCosts) {
         .and_then(|v| v.as_f64())
         .unwrap_or(0.0);
 
-    let Some(data) = span.data.value_mut() else {
-        return;
-    };
+    let data = span.data.get_or_insert_with(SpanData::default);
 
     // Extracts the response tokens per second
     if data.gen_ai_response_tokens_per_second.value().is_none()