feat: 更新主流大模型供应商目录

- 按云服务、Coding Plan、API 网关和本地模型分组

- 新增兼容模型发现与供应商协议校验

backend/pkg/agentcli/llm/anthropic.go

@@ -1,7 +1,7 @@
 // anthropic.go - Anthropic Messages 协议 LLM 客户端实现 // Logan.Li
 // 支持官方 Claude（api.anthropic.com）及智谱 GLM 编码套餐（open.bigmodel.cn/api/anthropic）等
 // 所有 Anthropic /v1/messages 兼容端点。与 openai.go 实现同一 Client 接口，由 client.go 按 Protocol 路由。
-// 鉴权统一用 Authorization: Bearer（智谱编码套餐实测通过；官方 Claude 较新版本亦支持）+ anthropic-version。
+// 同时发送 x-api-key 与 Bearer，兼容 Anthropic 官方和国内编码套餐网关。 // Logan.Li
 package llm
 
 import (
@@ -167,7 +167,10 @@ func buildTools(tools []Tool) []anthropicTool {
 // isAnthropicThinkingModel 判断是否为支持 thinking 的 Claude 模型 // Logan.Li
 func isAnthropicThinkingModel(model string) bool {
 	m := strings.ToLower(model)
-	return strings.Contains(m, "claude-3-7") || strings.Contains(m, "claude-3.7")
+	return strings.Contains(m, "claude-3-7") || strings.Contains(m, "claude-3.7") ||
+		strings.Contains(m, "claude-sonnet-4") || strings.Contains(m, "claude-opus-4") ||
+		strings.Contains(m, "claude-sonnet-5") || strings.Contains(m, "claude-opus-5") ||
+		strings.Contains(m, "claude-fable-5")
 }
 
 // buildRequest 构造 Anthropic 请求体（Complete 与 Stream 共用） // Logan.Li
@@ -209,6 +212,7 @@ func (c *AnthropicClient) newHTTPRequest(ctx context.Context, body []byte) (*htt
 	}
 	httpReq.Header.Set("Content-Type", "application/json")
 	httpReq.Header.Set("Authorization", "Bearer "+c.cfg.APIKey)
+	httpReq.Header.Set("x-api-key", c.cfg.APIKey)
 	httpReq.Header.Set("anthropic-version", "2023-06-01")
 	return httpReq, nil
 }
@@ -356,8 +360,8 @@ func (c *AnthropicClient) Stream(ctx context.Context, req *CompletionRequest, on
 		}
 	}
 	return &CompletionResponse{
-		Model:  c.cfg.Model,
+		Model:   c.cfg.Model,
 		Message: Message{Role: "assistant", Content: contentBuilder.String()},
-		Usage:  usage,
+		Usage:   usage,
 	}, nil
 }

backend/pkg/agentcli/llm/openai.go

@@ -49,10 +49,14 @@ func NewOpenAIClient(cfg Config) Client {
 // Name 返回客户端名称 // Logan.Li
 func (c *OpenAIClient) Name() string { return "openai-compatible(" + c.cfg.Model + ")" }
 
-// isOpenAIReasoningModel 判断是否为需要 reasoning_effort 的 OpenAI o-series 模型 // Logan.Li
+// isOpenAIReasoningModel 判断是否为使用新推理参数的 OpenAI 模型。 // Logan.Li
 func isOpenAIReasoningModel(model string) bool {
 	m := strings.ToLower(model)
-	return strings.Contains(m, "o1") || strings.Contains(m, "o3")
+	if index := strings.LastIndex(m, "/"); index >= 0 {
+		m = m[index+1:]
+	}
+	return strings.HasPrefix(m, "o1") || strings.HasPrefix(m, "o3") ||
+		strings.HasPrefix(m, "o4") || strings.HasPrefix(m, "gpt-5")
 }
 
 // isGenericReasoningModel 判断是否为通用推理开关模型（DeepSeek R1 等） // Logan.Li
@@ -61,33 +65,35 @@ func isGenericReasoningModel(model string) bool {
 	return strings.Contains(m, "reasoning") || strings.Contains(m, "r1") || strings.Contains(m, "deepseek-reasoner")
 }
 
-// applyReasoningParams 根据模型类型应用 reasoning 参数 // Logan.Li
+// applyReasoningParams 根据模型类型应用 token 与 reasoning 参数。 // Logan.Li
 func applyReasoningParams(payload *openaiRequest, enabled bool) {
-	if !enabled {
-		return
-	}
 	if isOpenAIReasoningModel(payload.Model) {
-		// OpenAI o-series 使用 reasoning_effort，temperature 必须固定为 1（不传） // Logan.Li
-		payload.ReasoningEffort = "medium"
+		// 新版 OpenAI 推理模型使用 max_completion_tokens 且不接收 temperature。 // Logan.Li
+		payload.MaxCompletionTokens = payload.MaxTokens
+		payload.MaxTokens = 0
 		payload.Temperature = 0
+		if enabled {
+			payload.ReasoningEffort = "medium"
+		}
 		return
 	}
-	if isGenericReasoningModel(payload.Model) {
+	if enabled && isGenericReasoningModel(payload.Model) {
 		payload.Reasoning = true
 	}
 }
 
 // openaiRequest OpenAI chat/completions 请求体 // Logan.Li
 type openaiRequest struct {
-	Model           string    `json:"model"`
-	Messages        []Message `json:"messages"`
-	Tools           []Tool    `json:"tools,omitempty"`
-	ToolChoice      string    `json:"tool_choice,omitempty"` // "auto" 让模型自主决定
-	Temperature     float64   `json:"temperature,omitempty"`
-	MaxTokens       int       `json:"max_tokens,omitempty"`
-	Stream          bool      `json:"stream,omitempty"`          // 流式补全（Stream 方法置 true）
-	ReasoningEffort string    `json:"reasoning_effort,omitempty"` // OpenAI o-series 推理强度 // Logan.Li
-	Reasoning       bool      `json:"reasoning,omitempty"`        // DeepSeek 等部分平台推理开关 // Logan.Li
+	Model               string    `json:"model"`
+	Messages            []Message `json:"messages"`
+	Tools               []Tool    `json:"tools,omitempty"`
+	ToolChoice          string    `json:"tool_choice,omitempty"` // "auto" 让模型自主决定
+	Temperature         float64   `json:"temperature,omitempty"`
+	MaxTokens           int       `json:"max_tokens,omitempty"`
+	MaxCompletionTokens int       `json:"max_completion_tokens,omitempty"` // 新版 OpenAI 推理模型的输出上限 // Logan.Li
+	Stream              bool      `json:"stream,omitempty"`                // 流式补全（Stream 方法置 true）
+	ReasoningEffort     string    `json:"reasoning_effort,omitempty"`      // OpenAI o-series 推理强度 // Logan.Li
+	Reasoning           bool      `json:"reasoning,omitempty"`             // DeepSeek 等部分平台推理开关 // Logan.Li
 }
 
 // openaiResponse OpenAI chat/completions 响应体 // Logan.Li
@@ -245,9 +251,9 @@ func (c *OpenAIClient) Stream(ctx context.Context, req *CompletionRequest, onDel
 		}
 	}
 	return &CompletionResponse{
-		Model:  c.cfg.Model,
+		Model:   c.cfg.Model,
 		Message: Message{Role: "assistant", Content: contentBuilder.String()},
-		Usage:  usage,
+		Usage:   usage,
 	}, nil
 }
 

backend/pkg/agentcli/llm/provider_compat_test.go

@@ -0,0 +1,39 @@
+package llm
+
+import "testing"
+
+func TestApplyReasoningParamsForModernOpenAIModels(t *testing.T) {
+	payload := openaiRequest{
+		Model:       "openai/gpt-5.4",
+		Temperature: 0.7,
+		MaxTokens:   8192,
+	}
+	applyReasoningParams(&payload, true)
+
+	if payload.Temperature != 0 || payload.MaxTokens != 0 {
+		t.Fatalf("unsupported parameters were not cleared: %+v", payload)
+	}
+	if payload.MaxCompletionTokens != 8192 || payload.ReasoningEffort != "medium" {
+		t.Fatalf("modern OpenAI parameters were not applied: %+v", payload)
+	}
+}
+
+func TestApplyReasoningParamsWithoutReasoningEffort(t *testing.T) {
+	payload := openaiRequest{Model: "gpt-5.4-mini", Temperature: 0.7, MaxTokens: 4096}
+	applyReasoningParams(&payload, false)
+
+	if payload.MaxCompletionTokens != 4096 || payload.ReasoningEffort != "" {
+		t.Fatalf("model compatibility parameters are incorrect: %+v", payload)
+	}
+}
+
+func TestAnthropicThinkingModels(t *testing.T) {
+	for _, model := range []string{"claude-3-7-sonnet", "claude-sonnet-4-6", "claude-opus-4-8", "claude-sonnet-5", "claude-fable-5"} {
+		if !isAnthropicThinkingModel(model) {
+			t.Fatalf("%q should support thinking parameters", model)
+		}
+	}
+	if isAnthropicThinkingModel("claude-haiku-4-5") {
+		t.Fatal("Claude Haiku should not receive thinking parameters")
+	}
+}