Remove deprecated Claude 3.5 models (#4590)

crates/goose/tests/agent.rs

@@ -244,21 +244,11 @@ mod tests {
         .await
     }
 
-    #[tokio::test]
-    async fn test_agent_with_anthropic() -> Result<()> {
-        run_test_with_config(TestConfig {
-            provider_type: ProviderType::Anthropic,
-            model: "claude-3-5-haiku-latest",
-            context_window: 200_000,
-        })
-        .await
-    }
-
     #[tokio::test]
     async fn test_agent_with_bedrock() -> Result<()> {
         run_test_with_config(TestConfig {
             provider_type: ProviderType::Bedrock,
-            model: "anthropic.claude-3-5-sonnet-20241022-v2:0",
+            model: "anthropic.claude-sonnet-4-20250514:0",
             context_window: 200_000,
         })
         .await
@@ -278,7 +268,7 @@ mod tests {
     async fn test_agent_with_databricks_bedrock() -> Result<()> {
         run_test_with_config(TestConfig {
             provider_type: ProviderType::Databricks,
-            model: "claude-3-5-sonnet-2",
+            model: "claude-sonnet-4",
             context_window: 200_000,
         })
         .await
@@ -338,7 +328,7 @@ mod tests {
     async fn test_agent_with_gcpvertexai() -> Result<()> {
         run_test_with_config(TestConfig {
             provider_type: ProviderType::GcpVertexAI,
-            model: "claude-3-5-sonnet-v2@20241022",
+            model: "claude-sonnet-4-20250514",
             context_window: 200_000,
         })
         .await

crates/goose/tests/pricing_integration_test.rs

@@ -18,11 +18,11 @@ async fn test_pricing_cache_performance() {
 
     // Test fetching pricing for common models (using actual model names from OpenRouter)
     let models = vec![
-        ("anthropic", "claude-3.5-sonnet"),
+        ("anthropic", "claude-sonnet-4"),
         ("openai", "gpt-4o"),
         ("openai", "gpt-4o-mini"),
         ("google", "gemini-flash-1.5"),
-        ("anthropic", "claude-sonnet-4"),
+        ("anthropic", "claude-opus-4"),
     ];
 
     // First fetch (potentially uncached or cache warming)
@@ -153,7 +153,7 @@ async fn run_pricing_refresh_test() -> Result<(), String> {
         .map_err(|e| format!("Failed to initialize pricing cache: {}", e))?;
 
     // Get initial pricing (using a model that actually exists)
-    let initial_pricing = get_model_pricing("anthropic", "claude-3.5-sonnet").await;
+    let initial_pricing = get_model_pricing("anthropic", "claude-sonnet-4").await;
     if initial_pricing.is_none() {
         return Err("Expected initial pricing but got None".to_string());
     }
@@ -167,7 +167,7 @@ async fn run_pricing_refresh_test() -> Result<(), String> {
     println!("Pricing refresh took: {:?}", refresh_duration);
 
     // Get pricing after refresh
-    let refreshed_pricing = get_model_pricing("anthropic", "claude-3.5-sonnet").await;
+    let refreshed_pricing = get_model_pricing("anthropic", "claude-sonnet-4").await;
     if refreshed_pricing.is_none() {
         return Err("Expected pricing after refresh but got None".to_string());
     }