feat: enable prompt caching and cache token tracking for google-vertex-anthropic (#20266)

Signed-off-by: Major Hayden <major@mhtx.net>

packages/opencode/test/provider/transform.test.ts

@@ -1792,6 +1792,58 @@ describe("ProviderTransform.message - cache control on gateway", () => {
       },
     })
   })
+
+  test("google-vertex-anthropic applies cache control", () => {
+    const model = createModel({
+      providerID: "google-vertex-anthropic",
+      api: {
+        id: "google-vertex-anthropic",
+        url: "https://us-central1-aiplatform.googleapis.com",
+        npm: "@ai-sdk/google-vertex/anthropic",
+      },
+      id: "claude-sonnet-4@20250514",
+    })
+    const msgs = [
+      {
+        role: "system",
+        content: "You are a helpful assistant",
+      },
+      {
+        role: "user",
+        content: "Hello",
+      },
+    ] as any[]
+
+    const result = ProviderTransform.message(msgs, model, {}) as any[]
+
+    expect(result[0].providerOptions).toEqual({
+      anthropic: {
+        cacheControl: {
+          type: "ephemeral",
+        },
+      },
+      openrouter: {
+        cacheControl: {
+          type: "ephemeral",
+        },
+      },
+      bedrock: {
+        cachePoint: {
+          type: "default",
+        },
+      },
+      openaiCompatible: {
+        cache_control: {
+          type: "ephemeral",
+        },
+      },
+      copilot: {
+        copilot_cache_control: {
+          type: "ephemeral",
+        },
+      },
+    })
+  })
 })
 
 describe("ProviderTransform.variants", () => {

packages/opencode/test/session/compaction.test.ts

@@ -1199,4 +1199,26 @@ describe("session.getUsage", () => {
       expect(result.tokens.total).toBe(1500)
     },
   )
+
+  test("extracts cache write tokens from vertex metadata key", () => {
+    const model = createModel({ context: 100_000, output: 32_000, npm: "@ai-sdk/google-vertex/anthropic" })
+    const result = Session.getUsage({
+      model,
+      usage: {
+        inputTokens: 1000,
+        outputTokens: 500,
+        totalTokens: 1500,
+        cachedInputTokens: 200,
+      },
+      metadata: {
+        vertex: {
+          cacheCreationInputTokens: 300,
+        },
+      },
+    })
+
+    expect(result.tokens.input).toBe(500)
+    expect(result.tokens.cache.read).toBe(200)
+    expect(result.tokens.cache.write).toBe(300)
+  })
 })