fix(claude): ensure max_tokens exceeds thinking budget for thinking models

Fixes an issue where Claude thinking models would return 400 errors when
the thinking.budget_tokens was greater than or equal to max_tokens.

Changes:
- Add MaxCompletionTokens: 128000 to all Claude thinking model definitions
- Add ensureMaxTokensForThinking() function in claude_executor.go that:
  - Checks if thinking is enabled with a budget_tokens value
  - Looks up the model's MaxCompletionTokens from the registry
  - Ensures max_tokens is set to at least the model's MaxCompletionTokens
  - Falls back to budget_tokens + 4000 buffer if registry lookup fails

This ensures Anthropic API constraint (max_tokens > thinking.budget_tokens)
is always satisfied when using extended thinking features.

Fixes: #339

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

internal/registry/model_definitions.go

@@ -30,6 +30,7 @@ func GetClaudeModels() []*ModelInfo {
 			OwnedBy:             "anthropic",
 			Type:                "claude",
 			DisplayName:         "Claude 4.5 Sonnet Thinking",
+			MaxCompletionTokens: 128000,
 			Thinking:            &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
 		},
 		{
@@ -39,6 +40,7 @@ func GetClaudeModels() []*ModelInfo {
 			OwnedBy:             "anthropic",
 			Type:                "claude",
 			DisplayName:         "Claude 4.5 Opus Thinking",
+			MaxCompletionTokens: 128000,
 			Thinking:            &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
 		},
 		{
@@ -48,6 +50,7 @@ func GetClaudeModels() []*ModelInfo {
 			OwnedBy:             "anthropic",
 			Type:                "claude",
 			DisplayName:         "Claude 4.5 Opus Thinking Low",
+			MaxCompletionTokens: 128000,
 			Thinking:            &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
 		},
 		{
@@ -57,6 +60,7 @@ func GetClaudeModels() []*ModelInfo {
 			OwnedBy:             "anthropic",
 			Type:                "claude",
 			DisplayName:         "Claude 4.5 Opus Thinking Medium",
+			MaxCompletionTokens: 128000,
 			Thinking:            &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
 		},
 		{
@@ -66,6 +70,7 @@ func GetClaudeModels() []*ModelInfo {
 			OwnedBy:             "anthropic",
 			Type:                "claude",
 			DisplayName:         "Claude 4.5 Opus Thinking High",
+			MaxCompletionTokens: 128000,
 			Thinking:            &ThinkingSupport{Min: 1024, Max: 100000, ZeroAllowed: false, DynamicAllowed: true},
 		},
 		{

internal/runtime/executor/claude_executor.go

@@ -17,6 +17,7 @@ import (
 	claudeauth "github.com/router-for-me/CLIProxyAPI/v6/internal/auth/claude"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/config"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/misc"
+	"github.com/router-for-me/CLIProxyAPI/v6/internal/registry"
 	"github.com/router-for-me/CLIProxyAPI/v6/internal/util"
 	cliproxyauth "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/auth"
 	cliproxyexecutor "github.com/router-for-me/CLIProxyAPI/v6/sdk/cliproxy/executor"
@@ -66,6 +67,9 @@ func (e *ClaudeExecutor) Execute(ctx context.Context, auth *cliproxyauth.Auth, r
 	}
 	body = applyPayloadConfig(e.cfg, req.Model, body)
 
+	// Ensure max_tokens > thinking.budget_tokens when thinking is enabled
+	body = ensureMaxTokensForThinking(req.Model, body)
+
 	// Extract betas from body and convert to header
 	var extraBetas []string
 	extraBetas, body = extractAndRemoveBetas(body)
@@ -165,6 +169,9 @@ func (e *ClaudeExecutor) ExecuteStream(ctx context.Context, auth *cliproxyauth.A
 	body = checkSystemInstructions(body)
 	body = applyPayloadConfig(e.cfg, req.Model, body)
 
+	// Ensure max_tokens > thinking.budget_tokens when thinking is enabled
+	body = ensureMaxTokensForThinking(req.Model, body)
+
 	// Extract betas from body and convert to header
 	var extraBetas []string
 	extraBetas, body = extractAndRemoveBetas(body)
@@ -447,6 +454,42 @@ func (e *ClaudeExecutor) injectThinkingConfig(modelName string, body []byte) []b
 	return body
 }
 
+// ensureMaxTokensForThinking ensures max_tokens > thinking.budget_tokens when thinking is enabled.
+// Anthropic API requires this constraint; violating it returns a 400 error.
+// This function should be called after all thinking configuration is finalized.
+// It looks up the model's MaxCompletionTokens from the registry to use as the cap.
+func ensureMaxTokensForThinking(modelName string, body []byte) []byte {
+	thinkingType := gjson.GetBytes(body, "thinking.type").String()
+	if thinkingType != "enabled" {
+		return body
+	}
+
+	budgetTokens := gjson.GetBytes(body, "thinking.budget_tokens").Int()
+	if budgetTokens <= 0 {
+		return body
+	}
+
+	maxTokens := gjson.GetBytes(body, "max_tokens").Int()
+
+	// Look up the model's max completion tokens from the registry
+	maxCompletionTokens := 0
+	if modelInfo := registry.GetGlobalRegistry().GetModelInfo(modelName); modelInfo != nil {
+		maxCompletionTokens = modelInfo.MaxCompletionTokens
+	}
+
+	// Fall back to budget + buffer if registry lookup fails or returns 0
+	const fallbackBuffer = 4000
+	requiredMaxTokens := budgetTokens + fallbackBuffer
+	if maxCompletionTokens > 0 {
+		requiredMaxTokens = int64(maxCompletionTokens)
+	}
+
+	if maxTokens < requiredMaxTokens {
+		body, _ = sjson.SetBytes(body, "max_tokens", requiredMaxTokens)
+	}
+	return body
+}
+
 func (e *ClaudeExecutor) resolveUpstreamModel(alias string, auth *cliproxyauth.Auth) string {
 	if alias == "" {
 		return ""