fix vercel gateway variants (#13541)

Co-authored-by: Benjamin Woodruff <[email protected]>"

2 files changed, 380 insertions, 10 deletions

diff --git a/packages/opencode/src/provider/transform.ts b/packages/opencode/src/provider/transform.ts
index 8091f731f..b4eb14a2f 100644
--- a/packages/opencode/src/provider/transform.ts
+++ b/packages/opencode/src/provider/transform.ts
@@ -171,7 +171,7 @@ export namespace ProviderTransform {
     return msgs
   }
 
-  function applyCaching(msgs: ModelMessage[], providerID: string): ModelMessage[] {
+  function applyCaching(msgs: ModelMessage[], model: Provider.Model): ModelMessage[] {
     const system = msgs.filter((msg) => msg.role === "system").slice(0, 2)
     const final = msgs.filter((msg) => msg.role !== "system").slice(-2)
 
@@ -194,7 +194,7 @@ export namespace ProviderTransform {
     }
 
     for (const msg of unique([...system, ...final])) {
-      const useMessageLevelOptions = providerID === "anthropic" || providerID.includes("bedrock")
+      const useMessageLevelOptions = model.providerID === "anthropic" || model.providerID.includes("bedrock")
       const shouldUseContentOptions = !useMessageLevelOptions && Array.isArray(msg.content) && msg.content.length > 0
 
       if (shouldUseContentOptions) {
@@ -253,14 +253,15 @@ export namespace ProviderTransform {
     msgs = unsupportedParts(msgs, model)
     msgs = normalizeMessages(msgs, model, options)
     if (
-      model.providerID === "anthropic" ||
-      model.api.id.includes("anthropic") ||
-      model.api.id.includes("claude") ||
-      model.id.includes("anthropic") ||
-      model.id.includes("claude") ||
-      model.api.npm === "@ai-sdk/anthropic"
+      (model.providerID === "anthropic" ||
+        model.api.id.includes("anthropic") ||
+        model.api.id.includes("claude") ||
+        model.id.includes("anthropic") ||
+        model.id.includes("claude") ||
+        model.api.npm === "@ai-sdk/anthropic") &&
+      model.api.npm !== "@ai-sdk/gateway"
     ) {
-      msgs = applyCaching(msgs, model.providerID)
+      msgs = applyCaching(msgs, model)
     }
 
     // Remap providerOptions keys from stored providerID to expected SDK key
@@ -363,8 +364,50 @@ export namespace ProviderTransform {
         if (!model.id.includes("gpt") && !model.id.includes("gemini-3")) return {}
         return Object.fromEntries(OPENAI_EFFORTS.map((effort) => [effort, { reasoning: { effort } }]))
 
-      // TODO: YOU CANNOT SET max_tokens if this is set!!!
       case "@ai-sdk/gateway":
+        if (model.id.includes("anthropic")) {
+          return {
+            high: {
+              thinking: {
+                type: "enabled",
+                budgetTokens: 16000,
+              },
+            },
+            max: {
+              thinking: {
+                type: "enabled",
+                budgetTokens: 31999,
+              },
+            },
+          }
+        }
+        if (model.id.includes("google")) {
+          if (id.includes("2.5")) {
+            return {
+              high: {
+                thinkingConfig: {
+                  includeThoughts: true,
+                  thinkingBudget: 16000,
+                },
+              },
+              max: {
+                thinkingConfig: {
+                  includeThoughts: true,
+                  thinkingBudget: 24576,
+                },
+              },
+            }
+          }
+          return Object.fromEntries(
+            ["low", "high"].map((effort) => [
+              effort,
+              {
+                includeThoughts: true,
+                thinkingLevel: effort,
+              },
+            ]),
+          )
+        }
         return Object.fromEntries(OPENAI_EFFORTS.map((effort) => [effort, { reasoningEffort: effort }]))
 
       case "@ai-sdk/github-copilot":
@@ -720,6 +763,12 @@ export namespace ProviderTransform {
       result["promptCacheKey"] = input.sessionID
     }
 
+    if (input.model.api.npm === "@ai-sdk/gateway") {
+      result["gateway"] = {
+        caching: "auto",
+      }
+    }
+
     return result
   }
 
@@ -754,6 +803,34 @@ export namespace ProviderTransform {
   }
 
   export function providerOptions(model: Provider.Model, options: { [x: string]: any }) {
+    if (model.api.npm === "@ai-sdk/gateway") {
+      // Gateway providerOptions are split across two namespaces:
+      // - `gateway`: gateway-native routing/caching controls
+      // - `<upstream slug>`: provider-specific model options (anthropic/openai/...)
+      // We keep `gateway` as-is and route every other top-level option under the
+      // model-derived upstream slug so variants/options can stay flat internally.
+      const i = model.api.id.indexOf("/")
+      const slug = i > 0 ? model.api.id.slice(0, i) : undefined
+      const gateway = options.gateway
+      const rest = Object.fromEntries(Object.entries(options).filter(([k]) => k !== "gateway"))
+      const has = Object.keys(rest).length > 0
+
+      const result: Record<string, any> = {}
+      if (gateway !== undefined) result.gateway = gateway
+
+      if (has) {
+        if (slug) {
+          result[slug] = rest
+        } else if (gateway && typeof gateway === "object" && !Array.isArray(gateway)) {
+          result.gateway = { ...gateway, ...rest }
+        } else {
+          result.gateway = rest
+        }
+      }
+
+      return result
+    }
+
     const key = sdkKey(model.api.npm) ?? model.providerID
     return { [key]: options }
   }
diff --git a/packages/opencode/test/provider/transform.test.ts b/packages/opencode/test/provider/transform.test.ts
index 02bb5278f..99acfcc81 100644
--- a/packages/opencode/test/provider/transform.test.ts
+++ b/packages/opencode/test/provider/transform.test.ts
@@ -175,6 +175,174 @@ describe("ProviderTransform.options - gpt-5 textVerbosity", () => {
   })
 })
 
+describe("ProviderTransform.options - gateway", () => {
+  const sessionID = "test-session-123"
+
+  const createModel = (id: string) =>
+    ({
+      id,
+      providerID: "vercel",
+      api: {
+        id,
+        url: "https://ai-gateway.vercel.sh/v3/ai",
+        npm: "@ai-sdk/gateway",
+      },
+      name: id,
+      capabilities: {
+        temperature: true,
+        reasoning: true,
+        attachment: true,
+        toolcall: true,
+        input: { text: true, audio: false, image: true, video: false, pdf: true },
+        output: { text: true, audio: false, image: false, video: false, pdf: false },
+        interleaved: false,
+      },
+      cost: {
+        input: 0.001,
+        output: 0.002,
+        cache: { read: 0.0001, write: 0.0002 },
+      },
+      limit: {
+        context: 200_000,
+        output: 8192,
+      },
+      status: "active",
+      options: {},
+      headers: {},
+      release_date: "2024-01-01",
+    }) as any
+
+  test("puts gateway defaults under gateway key", () => {
+    const model = createModel("anthropic/claude-sonnet-4")
+    const result = ProviderTransform.options({ model, sessionID, providerOptions: {} })
+    expect(result).toEqual({
+      gateway: {
+        caching: "auto",
+      },
+    })
+  })
+})
+
+describe("ProviderTransform.providerOptions", () => {
+  const createModel = (overrides: Partial<any> = {}) =>
+    ({
+      id: "test/test-model",
+      providerID: "test",
+      api: {
+        id: "test-model",
+        url: "https://api.test.com",
+        npm: "@ai-sdk/openai",
+      },
+      name: "Test Model",
+      capabilities: {
+        temperature: true,
+        reasoning: true,
+        attachment: true,
+        toolcall: true,
+        input: { text: true, audio: false, image: true, video: false, pdf: false },
+        output: { text: true, audio: false, image: false, video: false, pdf: false },
+        interleaved: false,
+      },
+      cost: {
+        input: 0.001,
+        output: 0.002,
+        cache: { read: 0.0001, write: 0.0002 },
+      },
+      limit: {
+        context: 200_000,
+        output: 64_000,
+      },
+      status: "active",
+      options: {},
+      headers: {},
+      release_date: "2024-01-01",
+      ...overrides,
+    }) as any
+
+  test("uses sdk key for non-gateway models", () => {
+    const model = createModel({
+      providerID: "my-bedrock",
+      api: {
+        id: "anthropic.claude-sonnet-4",
+        url: "https://bedrock.aws",
+        npm: "@ai-sdk/amazon-bedrock",
+      },
+    })
+
+    expect(ProviderTransform.providerOptions(model, { cachePoint: { type: "default" } })).toEqual({
+      bedrock: { cachePoint: { type: "default" } },
+    })
+  })
+
+  test("uses gateway model provider slug for gateway models", () => {
+    const model = createModel({
+      providerID: "vercel",
+      api: {
+        id: "anthropic/claude-sonnet-4",
+        url: "https://ai-gateway.vercel.sh/v3/ai",
+        npm: "@ai-sdk/gateway",
+      },
+    })
+
+    expect(ProviderTransform.providerOptions(model, { thinking: { type: "enabled", budgetTokens: 12_000 } })).toEqual({
+      anthropic: { thinking: { type: "enabled", budgetTokens: 12_000 } },
+    })
+  })
+
+  test("falls back to gateway key when gateway api id is unscoped", () => {
+    const model = createModel({
+      id: "anthropic/claude-sonnet-4",
+      providerID: "vercel",
+      api: {
+        id: "claude-sonnet-4",
+        url: "https://ai-gateway.vercel.sh/v3/ai",
+        npm: "@ai-sdk/gateway",
+      },
+    })
+
+    expect(ProviderTransform.providerOptions(model, { thinking: { type: "enabled", budgetTokens: 12_000 } })).toEqual({
+      gateway: { thinking: { type: "enabled", budgetTokens: 12_000 } },
+    })
+  })
+
+  test("splits gateway routing options from provider-specific options", () => {
+    const model = createModel({
+      providerID: "vercel",
+      api: {
+        id: "anthropic/claude-sonnet-4",
+        url: "https://ai-gateway.vercel.sh/v3/ai",
+        npm: "@ai-sdk/gateway",
+      },
+    })
+
+    expect(
+      ProviderTransform.providerOptions(model, {
+        gateway: { order: ["vertex", "anthropic"] },
+        thinking: { type: "enabled", budgetTokens: 12_000 },
+      }),
+    ).toEqual({
+      gateway: { order: ["vertex", "anthropic"] },
+      anthropic: { thinking: { type: "enabled", budgetTokens: 12_000 } },
+    } as any)
+  })
+
+  test("falls back to gateway key when model id has no provider slug", () => {
+    const model = createModel({
+      id: "claude-sonnet-4",
+      providerID: "vercel",
+      api: {
+        id: "claude-sonnet-4",
+        url: "https://ai-gateway.vercel.sh/v3/ai",
+        npm: "@ai-sdk/gateway",
+      },
+    })
+
+    expect(ProviderTransform.providerOptions(model, { reasoningEffort: "high" })).toEqual({
+      gateway: { reasoningEffort: "high" },
+    })
+  })
+})
+
 describe("ProviderTransform.schema - gemini array items", () => {
   test("adds missing items for array properties", () => {
     const geminiModel = {
@@ -1232,6 +1400,105 @@ describe("ProviderTransform.message - claude w/bedrock custom inference profile"
   })
 })
 
+describe("ProviderTransform.message - cache control on gateway", () => {
+  const createModel = (overrides: Partial<any> = {}) =>
+    ({
+      id: "anthropic/claude-sonnet-4",
+      providerID: "vercel",
+      api: {
+        id: "anthropic/claude-sonnet-4",
+        url: "https://ai-gateway.vercel.sh/v3/ai",
+        npm: "@ai-sdk/gateway",
+      },
+      name: "Claude Sonnet 4",
+      capabilities: {
+        temperature: true,
+        reasoning: true,
+        attachment: true,
+        toolcall: true,
+        input: { text: true, audio: false, image: true, video: false, pdf: true },
+        output: { text: true, audio: false, image: false, video: false, pdf: false },
+        interleaved: false,
+      },
+      cost: { input: 0.001, output: 0.002, cache: { read: 0.0001, write: 0.0002 } },
+      limit: { context: 200_000, output: 8192 },
+      status: "active",
+      options: {},
+      headers: {},
+      ...overrides,
+    }) as any
+
+  test("gateway does not set cache control for anthropic models", () => {
+    const model = createModel()
+    const msgs = [
+      {
+        role: "system",
+        content: [{ type: "text", text: "You are a helpful assistant" }],
+      },
+      {
+        role: "user",
+        content: "Hello",
+      },
+    ] as any[]
+
+    const result = ProviderTransform.message(msgs, model, {}) as any[]
+
+    expect(result[0].content[0].providerOptions).toBeUndefined()
+    expect(result[0].providerOptions).toBeUndefined()
+  })
+
+  test("non-gateway anthropic keeps existing cache control behavior", () => {
+    const model = createModel({
+      providerID: "anthropic",
+      api: {
+        id: "claude-sonnet-4",
+        url: "https://api.anthropic.com",
+        npm: "@ai-sdk/anthropic",
+      },
+    })
+    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", () => {
   const createMockModel = (overrides: Partial<any> = {}): any => ({
     id: "test/test-model",
@@ -1408,6 +1675,32 @@ describe("ProviderTransform.variants", () => {
   })
 
   describe("@ai-sdk/gateway", () => {
+    test("anthropic models return anthropic thinking options", () => {
+      const model = createMockModel({
+        id: "anthropic/claude-sonnet-4",
+        providerID: "gateway",
+        api: {
+          id: "anthropic/claude-sonnet-4",
+          url: "https://gateway.ai",
+          npm: "@ai-sdk/gateway",
+        },
+      })
+      const result = ProviderTransform.variants(model)
+      expect(Object.keys(result)).toEqual(["high", "max"])
+      expect(result.high).toEqual({
+        thinking: {
+          type: "enabled",
+          budgetTokens: 16000,
+        },
+      })
+      expect(result.max).toEqual({
+        thinking: {
+          type: "enabled",
+          budgetTokens: 31999,
+        },
+      })
+    })
+
     test("returns OPENAI_EFFORTS with reasoningEffort", () => {
       const model = createMockModel({
         id: "gateway/gateway-model",