fix: variant logic for anthropic models through openai compat endpoint (#11665)

2 files changed, 2 insertions, 170 deletions

diff --git a/packages/opencode/src/provider/transform.ts b/packages/opencode/src/provider/transform.ts
index 68220b90c..ded416e66 100644
--- a/packages/opencode/src/provider/transform.ts
+++ b/packages/opencode/src/provider/transform.ts
@@ -395,31 +395,6 @@ export namespace ProviderTransform {
       case "@ai-sdk/deepinfra":
       // https://v5.ai-sdk.dev/providers/ai-sdk-providers/deepinfra
       case "@ai-sdk/openai-compatible":
-        // When using openai-compatible SDK with Claude/Anthropic models,
-        // we must use snake_case (budget_tokens) as the SDK doesn't convert parameter names
-        // and the OpenAI-compatible API spec uses snake_case
-        if (
-          model.providerID === "anthropic" ||
-          model.api.id.includes("anthropic") ||
-          model.api.id.includes("claude") ||
-          model.id.includes("anthropic") ||
-          model.id.includes("claude")
-        ) {
-          return {
-            high: {
-              thinking: {
-                type: "enabled",
-                budget_tokens: 16000,
-              },
-            },
-            max: {
-              thinking: {
-                type: "enabled",
-                budget_tokens: 31999,
-              },
-            },
-          }
-        }
         return Object.fromEntries(WIDELY_SUPPORTED_EFFORTS.map((effort) => [effort, { reasoningEffort: effort }]))
 
       case "@ai-sdk/azure":
@@ -719,21 +694,9 @@ export namespace ProviderTransform {
     const modelCap = modelLimit || globalLimit
     const standardLimit = Math.min(modelCap, globalLimit)
 
-    // Handle thinking mode for @ai-sdk/anthropic, @ai-sdk/google-vertex/anthropic (budgetTokens)
-    // and @ai-sdk/openai-compatible with Claude (budget_tokens)
-    if (
-      npm === "@ai-sdk/anthropic" ||
-      npm === "@ai-sdk/google-vertex/anthropic" ||
-      npm === "@ai-sdk/openai-compatible"
-    ) {
+    if (npm === "@ai-sdk/anthropic" || npm === "@ai-sdk/google-vertex/anthropic") {
       const thinking = options?.["thinking"]
-      // Support both camelCase (for @ai-sdk/anthropic) and snake_case (for openai-compatible)
-      const budgetTokens =
-        typeof thinking?.["budgetTokens"] === "number"
-          ? thinking["budgetTokens"]
-          : typeof thinking?.["budget_tokens"] === "number"
-            ? thinking["budget_tokens"]
-            : 0
+      const budgetTokens = typeof thinking?.["budgetTokens"] === "number" ? thinking["budgetTokens"] : 0
       const enabled = thinking?.["type"] === "enabled"
       if (enabled && budgetTokens > 0) {
         // Return text tokens so that text + thinking <= model cap, preferring 32k text when possible.
diff --git a/packages/opencode/test/provider/transform.test.ts b/packages/opencode/test/provider/transform.test.ts
index cbb9ddbc4..8e28f1209 100644
--- a/packages/opencode/test/provider/transform.test.ts
+++ b/packages/opencode/test/provider/transform.test.ts
@@ -267,76 +267,6 @@ describe("ProviderTransform.maxOutputTokens", () => {
       expect(result).toBe(OUTPUT_TOKEN_MAX)
     })
   })
-
-  describe("openai-compatible with thinking options (snake_case)", () => {
-    test("returns 32k when budget_tokens + 32k <= modelLimit", () => {
-      const modelLimit = 100000
-      const options = {
-        thinking: {
-          type: "enabled",
-          budget_tokens: 10000,
-        },
-      }
-      const result = ProviderTransform.maxOutputTokens(
-        "@ai-sdk/openai-compatible",
-        options,
-        modelLimit,
-        OUTPUT_TOKEN_MAX,
-      )
-      expect(result).toBe(OUTPUT_TOKEN_MAX)
-    })
-
-    test("returns modelLimit - budget_tokens when budget_tokens + 32k > modelLimit", () => {
-      const modelLimit = 50000
-      const options = {
-        thinking: {
-          type: "enabled",
-          budget_tokens: 30000,
-        },
-      }
-      const result = ProviderTransform.maxOutputTokens(
-        "@ai-sdk/openai-compatible",
-        options,
-        modelLimit,
-        OUTPUT_TOKEN_MAX,
-      )
-      expect(result).toBe(20000)
-    })
-
-    test("returns 32k when thinking type is not enabled", () => {
-      const modelLimit = 100000
-      const options = {
-        thinking: {
-          type: "disabled",
-          budget_tokens: 10000,
-        },
-      }
-      const result = ProviderTransform.maxOutputTokens(
-        "@ai-sdk/openai-compatible",
-        options,
-        modelLimit,
-        OUTPUT_TOKEN_MAX,
-      )
-      expect(result).toBe(OUTPUT_TOKEN_MAX)
-    })
-
-    test("returns 32k when budget_tokens is 0", () => {
-      const modelLimit = 100000
-      const options = {
-        thinking: {
-          type: "enabled",
-          budget_tokens: 0,
-        },
-      }
-      const result = ProviderTransform.maxOutputTokens(
-        "@ai-sdk/openai-compatible",
-        options,
-        modelLimit,
-        OUTPUT_TOKEN_MAX,
-      )
-      expect(result).toBe(OUTPUT_TOKEN_MAX)
-    })
-  })
 })
 
 describe("ProviderTransform.schema - gemini array items", () => {
@@ -1564,67 +1494,6 @@ describe("ProviderTransform.variants", () => {
       expect(result.low).toEqual({ reasoningEffort: "low" })
       expect(result.high).toEqual({ reasoningEffort: "high" })
     })
-
-    test("Claude via LiteLLM returns thinking with snake_case budget_tokens", () => {
-      const model = createMockModel({
-        id: "anthropic/claude-sonnet-4-5",
-        providerID: "anthropic",
-        api: {
-          id: "claude-sonnet-4-5-20250929",
-          url: "http://localhost:4000",
-          npm: "@ai-sdk/openai-compatible",
-        },
-      })
-      const result = ProviderTransform.variants(model)
-      expect(Object.keys(result)).toEqual(["high", "max"])
-      expect(result.high).toEqual({
-        thinking: {
-          type: "enabled",
-          budget_tokens: 16000,
-        },
-      })
-      expect(result.max).toEqual({
-        thinking: {
-          type: "enabled",
-          budget_tokens: 31999,
-        },
-      })
-    })
-
-    test("Claude model (by model.id) via openai-compatible uses snake_case", () => {
-      const model = createMockModel({
-        id: "litellm/claude-3-opus",
-        providerID: "litellm",
-        api: {
-          id: "claude-3-opus-20240229",
-          url: "http://localhost:4000",
-          npm: "@ai-sdk/openai-compatible",
-        },
-      })
-      const result = ProviderTransform.variants(model)
-      expect(Object.keys(result)).toEqual(["high", "max"])
-      expect(result.high).toEqual({
-        thinking: {
-          type: "enabled",
-          budget_tokens: 16000,
-        },
-      })
-    })
-
-    test("Anthropic model (by model.api.id) via openai-compatible uses snake_case", () => {
-      const model = createMockModel({
-        id: "custom/my-model",
-        providerID: "custom",
-        api: {
-          id: "anthropic.claude-sonnet",
-          url: "http://localhost:4000",
-          npm: "@ai-sdk/openai-compatible",
-        },
-      })
-      const result = ProviderTransform.variants(model)
-      expect(Object.keys(result)).toEqual(["high", "max"])
-      expect(result.high.thinking.budget_tokens).toBe(16000)
-    })
   })
 
   describe("@ai-sdk/azure", () => {