@aituber-onair/chat 0.48.0 → 0.49.0

package/dist/umd/aituber-onair-chat.js

@@ -75,6 +75,7 @@ var AITuberOnAirChat = (() => {
     MODEL_ANTHROPIC_CLAUDE_4_5_HAIKU: () => MODEL_ANTHROPIC_CLAUDE_4_5_HAIKU,
     MODEL_ANTHROPIC_CLAUDE_HAIKU_LATEST: () => MODEL_ANTHROPIC_CLAUDE_HAIKU_LATEST,
     MODEL_ANTHROPIC_CLAUDE_OPUS_4: () => MODEL_ANTHROPIC_CLAUDE_OPUS_4,
+    MODEL_ANTHROPIC_CLAUDE_OPUS_5: () => MODEL_ANTHROPIC_CLAUDE_OPUS_5,
     MODEL_ANTHROPIC_CLAUDE_SONNET_4: () => MODEL_ANTHROPIC_CLAUDE_SONNET_4,
     MODEL_ANTHROPIC_CLAUDE_SONNET_LATEST: () => MODEL_ANTHROPIC_CLAUDE_SONNET_LATEST,
     MODEL_CLAUDE_3_5_HAIKU: () => MODEL_CLAUDE_3_5_HAIKU,
@@ -90,6 +91,7 @@ var AITuberOnAirChat = (() => {
     MODEL_CLAUDE_4_8_OPUS: () => MODEL_CLAUDE_4_8_OPUS,
     MODEL_CLAUDE_4_OPUS: () => MODEL_CLAUDE_4_OPUS,
     MODEL_CLAUDE_4_SONNET: () => MODEL_CLAUDE_4_SONNET,
+    MODEL_CLAUDE_5_OPUS: () => MODEL_CLAUDE_5_OPUS,
     MODEL_CLAUDE_5_SONNET: () => MODEL_CLAUDE_5_SONNET,
     MODEL_DEEPSEEK_CHAT: () => MODEL_DEEPSEEK_CHAT,
     MODEL_DEEPSEEK_REASONER: () => MODEL_DEEPSEEK_REASONER,
@@ -128,6 +130,8 @@ var AITuberOnAirChat = (() => {
     MODEL_GOOGLE_GEMINI_2_5_FLASH: () => MODEL_GOOGLE_GEMINI_2_5_FLASH,
     MODEL_GOOGLE_GEMINI_2_5_FLASH_LITE_PREVIEW_09_2025: () => MODEL_GOOGLE_GEMINI_2_5_FLASH_LITE_PREVIEW_09_2025,
     MODEL_GOOGLE_GEMINI_2_5_PRO: () => MODEL_GOOGLE_GEMINI_2_5_PRO,
+    MODEL_GOOGLE_GEMINI_3_5_FLASH_LITE: () => MODEL_GOOGLE_GEMINI_3_5_FLASH_LITE,
+    MODEL_GOOGLE_GEMINI_3_6_FLASH: () => MODEL_GOOGLE_GEMINI_3_6_FLASH,
     MODEL_GOOGLE_GEMINI_FLASH_LATEST: () => MODEL_GOOGLE_GEMINI_FLASH_LATEST,
     MODEL_GOOGLE_GEMINI_PRO_LATEST: () => MODEL_GOOGLE_GEMINI_PRO_LATEST,
     MODEL_GPT_4O: () => MODEL_GPT_4O,
@@ -160,6 +164,8 @@ var AITuberOnAirChat = (() => {
     MODEL_KIMI_K2_7_CODE: () => MODEL_KIMI_K2_7_CODE,
     MODEL_KIMI_K2_7_CODE_HIGHSPEED: () => MODEL_KIMI_K2_7_CODE_HIGHSPEED,
     MODEL_KIMI_K3: () => MODEL_KIMI_K3,
+    MODEL_KWAIPILOT_KAT_CODER_AIR_V2_5: () => MODEL_KWAIPILOT_KAT_CODER_AIR_V2_5,
+    MODEL_KWAIPILOT_KAT_CODER_PRO_V2_5: () => MODEL_KWAIPILOT_KAT_CODER_PRO_V2_5,
     MODEL_MINISTRAL_14B_2512: () => MODEL_MINISTRAL_14B_2512,
     MODEL_MINISTRAL_3B_2512: () => MODEL_MINISTRAL_3B_2512,
     MODEL_MINISTRAL_8B_2512: () => MODEL_MINISTRAL_8B_2512,
@@ -171,6 +177,7 @@ var AITuberOnAirChat = (() => {
     MODEL_MISTRAL_SMALL_LATEST: () => MODEL_MISTRAL_SMALL_LATEST,
     MODEL_MOONSHOTAI_KIMI_K2_5: () => MODEL_MOONSHOTAI_KIMI_K2_5,
     MODEL_MOONSHOTAI_KIMI_K2_7_CODE: () => MODEL_MOONSHOTAI_KIMI_K2_7_CODE,
+    MODEL_MOONSHOTAI_KIMI_K3: () => MODEL_MOONSHOTAI_KIMI_K3,
     MODEL_MOONSHOTAI_KIMI_LATEST: () => MODEL_MOONSHOTAI_KIMI_LATEST,
     MODEL_O1: () => MODEL_O1,
     MODEL_O1_MINI: () => MODEL_O1_MINI,
@@ -182,14 +189,20 @@ var AITuberOnAirChat = (() => {
     MODEL_OPENAI_GPT_5_1_CODEX: () => MODEL_OPENAI_GPT_5_1_CODEX,
     MODEL_OPENAI_GPT_5_5: () => MODEL_OPENAI_GPT_5_5,
     MODEL_OPENAI_GPT_5_5_PRO: () => MODEL_OPENAI_GPT_5_5_PRO,
+    MODEL_OPENAI_GPT_5_6_LUNA: () => MODEL_OPENAI_GPT_5_6_LUNA,
+    MODEL_OPENAI_GPT_5_6_SOL: () => MODEL_OPENAI_GPT_5_6_SOL,
+    MODEL_OPENAI_GPT_5_6_TERRA: () => MODEL_OPENAI_GPT_5_6_TERRA,
     MODEL_OPENAI_GPT_5_MINI: () => MODEL_OPENAI_GPT_5_MINI,
     MODEL_OPENAI_GPT_5_NANO: () => MODEL_OPENAI_GPT_5_NANO,
     MODEL_OPENAI_GPT_LATEST: () => MODEL_OPENAI_GPT_LATEST,
     MODEL_OPENAI_GPT_MINI_LATEST: () => MODEL_OPENAI_GPT_MINI_LATEST,
     MODEL_OPENROUTER_AUTO: () => MODEL_OPENROUTER_AUTO,
+    MODEL_OPENROUTER_AUTO_BETA: () => MODEL_OPENROUTER_AUTO_BETA,
     MODEL_OPENROUTER_FUSION: () => MODEL_OPENROUTER_FUSION,
     MODEL_PLAMO_2_2_PRIME: () => MODEL_PLAMO_2_2_PRIME,
     MODEL_PLAMO_3_0_PRIME: () => MODEL_PLAMO_3_0_PRIME,
+    MODEL_XAI_GROK_4_5: () => MODEL_XAI_GROK_4_5,
+    MODEL_XAI_GROK_LATEST: () => MODEL_XAI_GROK_LATEST,
     MODEL_ZAI_GLM_4_5_AIR: () => MODEL_ZAI_GLM_4_5_AIR,
     MODEL_ZAI_GLM_4_5_AIR_FREE: () => MODEL_ZAI_GLM_4_5_AIR_FREE,
     MODEL_ZAI_GLM_4_7_FLASH: () => MODEL_ZAI_GLM_4_7_FLASH,
@@ -229,13 +242,17 @@ var AITuberOnAirChat = (() => {
     allowsReasoningNone: () => allowsReasoningNone,
     allowsReasoningXHigh: () => allowsReasoningXHigh,
     buildOpenAICompatibleTools: () => buildOpenAICompatibleTools,
+    getClaudeSupportedReasoningEfforts: () => getClaudeSupportedReasoningEfforts,
+    getDefaultClaudeReasoningEffort: () => getDefaultClaudeReasoningEffort,
     getDefaultGeminiReasoningEffort: () => getDefaultGeminiReasoningEffort,
+    getDefaultKimiReasoningEffort: () => getDefaultKimiReasoningEffort,
     getDefaultReasoningEffortForGPT5Model: () => getDefaultReasoningEffortForGPT5Model,
     getDefaultXaiReasoningEffort: () => getDefaultXaiReasoningEffort,
     getGeminiSupportedReasoningEfforts: () => getGeminiSupportedReasoningEfforts,
     getKimiSupportedReasoningEfforts: () => getKimiSupportedReasoningEfforts,
     getMaxTokensForResponseLength: () => getMaxTokensForResponseLength,
     installGASFetch: () => installGASFetch,
+    isClaudeReasoningEffortModel: () => isClaudeReasoningEffortModel,
     isGPT5Model: () => isGPT5Model,
     isGeminiReasoningEffortModel: () => isGeminiReasoningEffortModel,
     isKimiReasoningEffortModel: () => isKimiReasoningEffortModel,
@@ -252,6 +269,7 @@ var AITuberOnAirChat = (() => {
     isXaiVisionModel: () => isXaiVisionModel,
     isZaiToolStreamModel: () => isZaiToolStreamModel,
     isZaiVisionModel: () => isZaiVisionModel,
+    normalizeClaudeReasoningEffort: () => normalizeClaudeReasoningEffort,
     normalizeGeminiReasoningEffort: () => normalizeGeminiReasoningEffort,
     normalizeXaiReasoningEffort: () => normalizeXaiReasoningEffort,
     parseOpenAICompatibleOneShot: () => parseOpenAICompatibleOneShot,
@@ -463,6 +481,30 @@ var AITuberOnAirChat = (() => {
   var MODEL_CLAUDE_4_7_OPUS = "claude-opus-4-7";
   var MODEL_CLAUDE_4_8_OPUS = "claude-opus-4-8";
   var MODEL_CLAUDE_5_SONNET = "claude-sonnet-5";
+  var MODEL_CLAUDE_5_OPUS = "claude-opus-5";
+  var CLAUDE_STANDARD_REASONING_EFFORTS = [
+    "low",
+    "medium",
+    "high"
+  ];
+  var CLAUDE_MAX_REASONING_EFFORTS = [
+    ...CLAUDE_STANDARD_REASONING_EFFORTS,
+    "max"
+  ];
+  var CLAUDE_XHIGH_REASONING_EFFORTS = [
+    ...CLAUDE_STANDARD_REASONING_EFFORTS,
+    "xhigh",
+    "max"
+  ];
+  var CLAUDE_REASONING_EFFORTS_BY_MODEL = {
+    [MODEL_CLAUDE_4_5_OPUS]: CLAUDE_STANDARD_REASONING_EFFORTS,
+    [MODEL_CLAUDE_4_6_SONNET]: CLAUDE_MAX_REASONING_EFFORTS,
+    [MODEL_CLAUDE_4_6_OPUS]: CLAUDE_MAX_REASONING_EFFORTS,
+    [MODEL_CLAUDE_4_7_OPUS]: CLAUDE_XHIGH_REASONING_EFFORTS,
+    [MODEL_CLAUDE_4_8_OPUS]: CLAUDE_XHIGH_REASONING_EFFORTS,
+    [MODEL_CLAUDE_5_SONNET]: CLAUDE_XHIGH_REASONING_EFFORTS,
+    [MODEL_CLAUDE_5_OPUS]: CLAUDE_XHIGH_REASONING_EFFORTS
+  };
   var CLAUDE_VISION_SUPPORTED_MODELS = [
     MODEL_CLAUDE_3_HAIKU,
     MODEL_CLAUDE_4_SONNET,
@@ -474,19 +516,47 @@ var AITuberOnAirChat = (() => {
     MODEL_CLAUDE_4_6_OPUS,
     MODEL_CLAUDE_4_7_OPUS,
     MODEL_CLAUDE_4_8_OPUS,
-    MODEL_CLAUDE_5_SONNET
+    MODEL_CLAUDE_5_SONNET,
+    MODEL_CLAUDE_5_OPUS
   ];
+  function getClaudeSupportedReasoningEfforts(model) {
+    return CLAUDE_REASONING_EFFORTS_BY_MODEL[model] ?? [];
+  }
+  function isClaudeReasoningEffortModel(model) {
+    return getClaudeSupportedReasoningEfforts(model).length > 0;
+  }
+  function getDefaultClaudeReasoningEffort(model) {
+    return isClaudeReasoningEffortModel(model) ? "high" : void 0;
+  }
+  function normalizeClaudeReasoningEffort(model, effort) {
+    const supported = getClaudeSupportedReasoningEfforts(model);
+    if (supported.length === 0) {
+      return void 0;
+    }
+    if (!effort) {
+      return getDefaultClaudeReasoningEffort(model);
+    }
+    if (supported.includes(effort)) {
+      return effort;
+    }
+    return "high";
+  }
 
   // src/constants/openrouter.ts
   var ENDPOINT_OPENROUTER_API = "https://openrouter.ai/api/v1/chat/completions";
   var MODEL_OPENROUTER_AUTO = "openrouter/auto";
+  var MODEL_OPENROUTER_AUTO_BETA = "openrouter/auto-beta";
   var MODEL_OPENROUTER_FUSION = "openrouter/fusion";
   var MODEL_GPT_OSS_20B_FREE = "openai/gpt-oss-20b:free";
+  var MODEL_MOONSHOTAI_KIMI_K3 = "moonshotai/kimi-k3";
   var MODEL_MOONSHOTAI_KIMI_K2_7_CODE = "moonshotai/kimi-k2.7-code";
   var MODEL_MOONSHOTAI_KIMI_K2_5 = "moonshotai/kimi-k2.5";
   var MODEL_MOONSHOTAI_KIMI_LATEST = "~moonshotai/kimi-latest";
   var MODEL_OPENAI_GPT_LATEST = "~openai/gpt-latest";
   var MODEL_OPENAI_GPT_MINI_LATEST = "~openai/gpt-mini-latest";
+  var MODEL_OPENAI_GPT_5_6_SOL = "openai/gpt-5.6-sol";
+  var MODEL_OPENAI_GPT_5_6_TERRA = "openai/gpt-5.6-terra";
+  var MODEL_OPENAI_GPT_5_6_LUNA = "openai/gpt-5.6-luna";
   var MODEL_OPENAI_GPT_5_5_PRO = "openai/gpt-5.5-pro";
   var MODEL_OPENAI_GPT_5_5 = "openai/gpt-5.5";
   var MODEL_OPENAI_GPT_5_1_CHAT = "openai/gpt-5.1-chat";
@@ -498,6 +568,7 @@ var AITuberOnAirChat = (() => {
   var MODEL_OPENAI_GPT_4_1_NANO = "openai/gpt-4.1-nano";
   var MODEL_ANTHROPIC_CLAUDE_SONNET_LATEST = "~anthropic/claude-sonnet-latest";
   var MODEL_ANTHROPIC_CLAUDE_HAIKU_LATEST = "~anthropic/claude-haiku-latest";
+  var MODEL_ANTHROPIC_CLAUDE_OPUS_5 = "anthropic/claude-opus-5";
   var MODEL_ANTHROPIC_CLAUDE_OPUS_4 = "anthropic/claude-opus-4";
   var MODEL_ANTHROPIC_CLAUDE_SONNET_4 = "anthropic/claude-sonnet-4";
   var MODEL_ANTHROPIC_CLAUDE_3_7_SONNET = "anthropic/claude-3.7-sonnet";
@@ -505,6 +576,8 @@ var AITuberOnAirChat = (() => {
   var MODEL_ANTHROPIC_CLAUDE_4_5_HAIKU = "anthropic/claude-haiku-4.5";
   var MODEL_GOOGLE_GEMINI_PRO_LATEST = "~google/gemini-pro-latest";
   var MODEL_GOOGLE_GEMINI_FLASH_LATEST = "~google/gemini-flash-latest";
+  var MODEL_GOOGLE_GEMINI_3_6_FLASH = "google/gemini-3.6-flash";
+  var MODEL_GOOGLE_GEMINI_3_5_FLASH_LITE = "google/gemini-3.5-flash-lite";
   var MODEL_GOOGLE_GEMINI_2_5_PRO = "google/gemini-2.5-pro";
   var MODEL_GOOGLE_GEMINI_2_5_FLASH = "google/gemini-2.5-flash";
   var MODEL_GOOGLE_GEMINI_2_5_FLASH_LITE_PREVIEW_09_2025 = "google/gemini-2.5-flash-lite-preview-09-2025";
@@ -512,14 +585,23 @@ var AITuberOnAirChat = (() => {
   var MODEL_ZAI_GLM_4_7_FLASH = "z-ai/glm-4.7-flash";
   var MODEL_ZAI_GLM_4_5_AIR = "z-ai/glm-4.5-air";
   var MODEL_ZAI_GLM_4_5_AIR_FREE = "z-ai/glm-4.5-air:free";
+  var MODEL_KWAIPILOT_KAT_CODER_AIR_V2_5 = "kwaipilot/kat-coder-air-v2.5";
+  var MODEL_KWAIPILOT_KAT_CODER_PRO_V2_5 = "kwaipilot/kat-coder-pro-v2.5";
+  var MODEL_XAI_GROK_LATEST = "~x-ai/grok-latest";
+  var MODEL_XAI_GROK_4_5 = "x-ai/grok-4.5";
   var OPENROUTER_FREE_MODELS = [
     MODEL_GPT_OSS_20B_FREE,
     MODEL_ZAI_GLM_4_5_AIR_FREE
   ];
   var OPENROUTER_VISION_SUPPORTED_MODELS = [
+    MODEL_OPENROUTER_AUTO_BETA,
+    MODEL_MOONSHOTAI_KIMI_K3,
     MODEL_MOONSHOTAI_KIMI_LATEST,
     MODEL_OPENAI_GPT_LATEST,
     MODEL_OPENAI_GPT_MINI_LATEST,
+    MODEL_OPENAI_GPT_5_6_SOL,
+    MODEL_OPENAI_GPT_5_6_TERRA,
+    MODEL_OPENAI_GPT_5_6_LUNA,
     MODEL_OPENAI_GPT_5_5_PRO,
     MODEL_OPENAI_GPT_5_5,
     MODEL_OPENAI_GPT_5_1_CHAT,
@@ -531,17 +613,22 @@ var AITuberOnAirChat = (() => {
     MODEL_OPENAI_GPT_4_1_NANO,
     MODEL_ANTHROPIC_CLAUDE_SONNET_LATEST,
     MODEL_ANTHROPIC_CLAUDE_HAIKU_LATEST,
+    MODEL_ANTHROPIC_CLAUDE_OPUS_5,
     MODEL_ANTHROPIC_CLAUDE_OPUS_4,
     MODEL_ANTHROPIC_CLAUDE_SONNET_4,
     MODEL_ANTHROPIC_CLAUDE_3_7_SONNET,
     MODEL_ANTHROPIC_CLAUDE_4_5_HAIKU,
     MODEL_GOOGLE_GEMINI_PRO_LATEST,
     MODEL_GOOGLE_GEMINI_FLASH_LATEST,
+    MODEL_GOOGLE_GEMINI_3_6_FLASH,
+    MODEL_GOOGLE_GEMINI_3_5_FLASH_LITE,
     MODEL_GOOGLE_GEMINI_2_5_PRO,
     MODEL_GOOGLE_GEMINI_2_5_FLASH,
     MODEL_GOOGLE_GEMINI_2_5_FLASH_LITE_PREVIEW_09_2025,
     MODEL_MOONSHOTAI_KIMI_K2_7_CODE,
-    MODEL_MOONSHOTAI_KIMI_K2_5
+    MODEL_MOONSHOTAI_KIMI_K2_5,
+    MODEL_XAI_GROK_LATEST,
+    MODEL_XAI_GROK_4_5
   ];
   var OPENROUTER_FREE_RATE_LIMIT_PER_MINUTE = 20;
   var OPENROUTER_FREE_DAILY_LIMIT_LOW_CREDITS = 50;
@@ -632,7 +719,7 @@ var AITuberOnAirChat = (() => {
   var MODEL_KIMI_K2_6 = "kimi-k2.6";
   var MODEL_KIMI_K2_5 = "kimi-k2.5";
   var KIMI_REASONING_EFFORTS_BY_MODEL = {
-    [MODEL_KIMI_K3]: ["max"]
+    [MODEL_KIMI_K3]: ["low", "high", "max"]
   };
   var KIMI_VISION_SUPPORTED_MODELS = [
     MODEL_KIMI_K3,
@@ -654,6 +741,9 @@ var AITuberOnAirChat = (() => {
   function getKimiSupportedReasoningEfforts(model) {
     return KIMI_REASONING_EFFORTS_BY_MODEL[model] ?? [];
   }
+  function getDefaultKimiReasoningEffort(model) {
+    return model === MODEL_KIMI_K3 ? "max" : void 0;
+  }
   function isKimiReasoningEffortModel(model) {
     return getKimiSupportedReasoningEfforts(model).length > 0;
   }
@@ -1090,6 +1180,15 @@ If it's in another language, summarize in that language.
       return "";
     }).join("");
   };
+  var throwIfApiError = (data) => {
+    const choice = data.choices?.[0];
+    const error = data.error ?? choice?.error;
+    if (!error && choice?.finish_reason !== "error") {
+      return;
+    }
+    const message = typeof error === "string" ? error : typeof error?.message === "string" ? error.message : "The provider ended the response with an error.";
+    throw new Error(`Provider response error: ${message}`);
+  };
   var forEachSsePayload = async (res, onPayload) => {
     const reader = res.body?.getReader();
     if (!reader) {
@@ -1122,6 +1221,9 @@ If it's in another language, summarize in that language.
     await forEachSsePayload(res, (payload) => {
       const json = parseJsonPayload(payload, options.onJsonError);
       if (!json) return;
+      if (options.throwOnApiError) {
+        throwIfApiError(json);
+      }
       const content = extractTextContent(json.choices?.[0]?.delta?.content);
       if (content) {
         onPartial(content);
@@ -1141,6 +1243,9 @@ If it's in another language, summarize in that language.
     await forEachSsePayload(res, (payload) => {
       const json = parseJsonPayload(payload, options.onJsonError);
       if (!json) return;
+      if (options.throwOnApiError) {
+        throwIfApiError(json);
+      }
       const choice = json.choices?.[0];
       if (typeof choice?.finish_reason === "string") {
         finishReason = choice.finish_reason;
@@ -1208,6 +1313,9 @@ If it's in another language, summarize in that language.
     };
   }
   function parseOpenAICompatibleOneShot(data, options = {}) {
+    if (options.throwOnApiError) {
+      throwIfApiError(data);
+    }
     const choice = data?.choices?.[0];
     const blocks = [];
     const content = extractTextContent(choice?.message?.content);
@@ -1515,11 +1623,17 @@ If it's in another language, summarize in that language.
   function convertMessagesToClaudeFormat(messages) {
     return messages.map((msg) => ({
       role: mapRoleToClaude(msg.role),
-      content: msg.content
+      content: msg.provider_content ?? msg.content
     }));
   }
   function convertVisionMessagesToClaudeFormat(messages) {
     return messages.map((msg) => {
+      if (msg.provider_content) {
+        return {
+          role: mapRoleToClaude(msg.role),
+          content: msg.provider_content
+        };
+      }
       if (typeof msg.content === "string") {
         return {
           role: mapRoleToClaude(msg.role),
@@ -1610,7 +1724,7 @@ If it's in another language, summarize in that language.
      * @param mcpServers Array of MCP server configurations (optional)
      * @throws Error if the vision model doesn't support vision capabilities
      */
-    constructor(apiKey, model = MODEL_CLAUDE_4_5_HAIKU, visionModel = MODEL_CLAUDE_4_5_HAIKU, tools = [], mcpServers = [], responseLength) {
+    constructor(apiKey, model = MODEL_CLAUDE_4_5_HAIKU, visionModel = MODEL_CLAUDE_4_5_HAIKU, tools = [], mcpServers = [], responseLength, reasoningEffort) {
       /** Provider name */
       this.provider = "claude";
       this.apiKey = apiKey;
@@ -1619,6 +1733,7 @@ If it's in another language, summarize in that language.
       this.tools = tools;
       this.mcpServers = mcpServers;
       this.responseLength = responseLength;
+      this.reasoningEffort = reasoningEffort;
       if (!CLAUDE_VISION_SUPPORTED_MODELS.includes(this.visionModel)) {
         throw new Error(
           `Model ${this.visionModel} does not support vision capabilities.`
@@ -1728,6 +1843,9 @@ If it's in another language, summarize in that language.
         stream,
         max_tokens: maxTokens !== void 0 ? maxTokens : getMaxTokensForResponseLength(this.responseLength)
       };
+      if (this.reasoningEffort !== void 0 && getClaudeSupportedReasoningEfforts(model).includes(this.reasoningEffort)) {
+        body.output_config = { effort: this.reasoningEffort };
+      }
       if (this.tools.length) {
         body.tools = this.tools.map((t) => ({
           name: t.name,
@@ -1765,6 +1883,7 @@ If it's in another language, summarize in that language.
       const reader = res.body.getReader();
       const dec = new TextDecoder();
       const textBlocks = [];
+      const providerBlocks = /* @__PURE__ */ new Map();
       const toolCalls = /* @__PURE__ */ new Map();
       let buf = "";
       while (true) {
@@ -1779,9 +1898,28 @@ If it's in another language, summarize in that language.
           const payload = line.slice(5).trim();
           if (payload === "[DONE]") break;
           const ev = JSON.parse(payload);
-          if (ev.type === "content_block_delta" && ev.delta?.text) {
+          if (ev.type === "content_block_start" && ev.content_block && typeof ev.index === "number") {
+            providerBlocks.set(ev.index, { ...ev.content_block });
+          }
+          if (ev.type === "content_block_delta" && ev.delta?.type === "text_delta") {
             onPartial(ev.delta.text);
             textBlocks.push({ type: "text", text: ev.delta.text });
+            const block = providerBlocks.get(ev.index);
+            if (block?.type === "text") {
+              block.text = `${block.text ?? ""}${ev.delta.text}`;
+            }
+          }
+          if (ev.type === "content_block_delta" && ev.delta?.type === "thinking_delta") {
+            const block = providerBlocks.get(ev.index);
+            if (block?.type === "thinking") {
+              block.thinking = `${block.thinking ?? ""}${ev.delta.thinking ?? ""}`;
+            }
+          }
+          if (ev.type === "content_block_delta" && ev.delta?.type === "signature_delta") {
+            const block = providerBlocks.get(ev.index);
+            if (block?.type === "thinking") {
+              block.signature = `${block.signature ?? ""}${ev.delta.signature ?? ""}`;
+            }
           }
           if (ev.type === "content_block_start" && ev.content_block?.type === "tool_use") {
             toolCalls.set(ev.index, {
@@ -1817,20 +1955,25 @@ If it's in another language, summarize in that language.
           }
           if (ev.type === "content_block_stop" && toolCalls.has(ev.index)) {
             const { id, name, args, server_name } = toolCalls.get(ev.index);
+            const input = JSON.parse(args || "{}");
+            const providerBlock = providerBlocks.get(ev.index);
+            if (providerBlock) {
+              providerBlock.input = input;
+            }
             if (server_name) {
               textBlocks.push({
                 type: "mcp_tool_use",
                 id,
                 name,
                 server_name,
-                input: JSON.parse(args || "{}")
+                input
               });
             } else {
               textBlocks.push({
                 type: "tool_use",
                 id,
                 name,
-                input: JSON.parse(args || "{}")
+                input
               });
             }
             toolCalls.delete(ev.index);
@@ -1839,6 +1982,7 @@ If it's in another language, summarize in that language.
       }
       return {
         blocks: textBlocks,
+        providerContent: [...providerBlocks.entries()].sort(([a], [b]) => a - b).map(([, block]) => block),
         stop_reason: textBlocks.some(
           (b) => b.type === "tool_use" || b.type === "mcp_tool_use"
         ) ? "tool_use" : "end"
@@ -1885,6 +2029,7 @@ If it's in another language, summarize in that language.
       });
       return {
         blocks,
+        providerContent: (data.content ?? []).map((block) => ({ ...block })),
         stop_reason: blocks.some(
           (b) => b.type === "tool_use" || b.type === "mcp_tool_use"
         ) ? "tool_use" : "end"
@@ -1936,7 +2081,12 @@ If it's in another language, summarize in that language.
       );
       return {
         blocks: standardBlocks,
-        stop_reason: completion.stop_reason
+        stop_reason: completion.stop_reason,
+        assistant_message: {
+          role: "assistant",
+          content: standardBlocks.filter((block) => block.type === "text").map((block) => block.text).join(""),
+          provider_content: completion.providerContent
+        }
       };
     }
   };
@@ -1949,21 +2099,30 @@ If it's in another language, summarize in that language.
      * @returns ClaudeChatService instance
      */
     createChatService(options) {
+      const model = options.model || this.getDefaultModel();
+      const supportedReasoningEfforts = getClaudeSupportedReasoningEfforts(model);
+      if (options.reasoning_effort !== void 0 && !supportedReasoningEfforts.includes(options.reasoning_effort)) {
+        const supportedMessage = supportedReasoningEfforts.length > 0 ? `Supported values: ${supportedReasoningEfforts.join(", ")}.` : "This model does not expose configurable effort.";
+        throw new Error(
+          `Model ${model} does not support Claude reasoning_effort: ${options.reasoning_effort}. ${supportedMessage}`
+        );
+      }
       const visionModel = resolveVisionModel({
-        model: options.model,
+        model,
         visionModel: options.visionModel,
         defaultModel: this.getDefaultModel(),
         defaultVisionModel: this.getDefaultModel(),
-        supportsVisionForModel: (model) => this.supportsVisionForModel(model),
+        supportsVisionForModel: (model2) => this.supportsVisionForModel(model2),
         validate: "resolved"
       });
       return new ClaudeChatService(
         options.apiKey,
-        options.model || this.getDefaultModel(),
+        model,
         visionModel,
         options.tools ?? [],
         options.mcpServers ?? [],
-        options.responseLength
+        options.responseLength,
+        options.reasoning_effort
       );
     }
     /**
@@ -1989,6 +2148,7 @@ If it's in another language, summarize in that language.
         MODEL_CLAUDE_4_7_OPUS,
         MODEL_CLAUDE_4_8_OPUS,
         MODEL_CLAUDE_5_SONNET,
+        MODEL_CLAUDE_5_OPUS,
         MODEL_CLAUDE_3_HAIKU
       ];
     }
@@ -3567,8 +3727,14 @@ If it's in another language, summarize in that language.
   var GeminiNanoChatService = class {
     constructor(options = {}) {
       this.provider = "gemini-nano";
-      this.expectedInputLanguages = options.expectedInputLanguages ?? ["ja"];
+      this.expectedInputLanguages = options.expectedInputLanguages ?? [
+        "ja",
+        "en"
+      ];
       this.expectedOutputLanguages = options.expectedOutputLanguages ?? ["ja"];
+      this.initialPrompts = (options.initialPrompts ?? []).map((prompt) => ({
+        ...prompt
+      }));
       this._responseLength = options.responseLength;
     }
     getModel() {
@@ -3613,7 +3779,8 @@ If it's in another language, summarize in that language.
           "Gemini Nano is not available in this environment. Chrome 138+ with Prompt API enabled is required."
         );
       }
-      const availability = await api.availability();
+      const modelOptions = this.getModelOptions();
+      const availability = await api.availability(modelOptions);
       if (availability !== "available" && availability !== "downloadable") {
         throw new Error(
           `Gemini Nano Prompt API is not ready in this environment. LanguageModel.availability() returned "${availability}". Expected "available" or "downloadable".`
@@ -3621,15 +3788,17 @@ If it's in another language, summarize in that language.
       }
       const systemMessages = messages.filter((m) => m.role === "system");
       const systemPrompt = systemMessages.map((m) => m.content).join("\n");
-      const conversationMessages = messages.filter((m) => m.role !== "system").slice(-GEMINI_NANO_MAX_CONTEXT_MESSAGES);
-      const lastUserMessage = [...conversationMessages].reverse().find((m) => m.role === "user");
-      if (!lastUserMessage) {
+      const conversationMessages = messages.filter((m) => m.role === "user" || m.role === "assistant").slice(-GEMINI_NANO_MAX_CONTEXT_MESSAGES);
+      const lastUserMessageIndex = this.findLastUserMessageIndex(conversationMessages);
+      if (lastUserMessageIndex === -1) {
         throw new Error("No user message found in the provided messages.");
       }
+      const lastUserMessage = conversationMessages[lastUserMessageIndex];
       const session = await this.createSession(
         api,
         systemPrompt,
-        conversationMessages
+        conversationMessages.slice(0, lastUserMessageIndex),
+        modelOptions
       );
       try {
         return await session.prompt(lastUserMessage.content);
@@ -3641,25 +3810,55 @@ If it's in another language, summarize in that language.
       }
     }
     /**
-     * Create a new LanguageModel session with system prompt and context history.
-     * Context history (excluding the last user message) is embedded in the system prompt.
+     * Create a new LanguageModel session with structured initial prompts.
+     * System instructions, few-shot examples, and conversation history retain
+     * their roles instead of being flattened into a single string.
      */
-    async createSession(api, systemPrompt, contextHistory) {
-      let prompt = this.buildSystemPrompt(systemPrompt);
-      const historyMessages = contextHistory.slice(0, -1);
-      if (historyMessages.length > 0) {
-        const history = historyMessages.map((m) => `${m.role === "user" ? "User" : "Assistant"}: ${m.content}`).join("\n");
-        prompt += "\n\nThe following is the prior conversation history. Use it as context for your response:\n" + history;
-      }
-      return api.create({
-        systemPrompt: prompt,
+    async createSession(api, systemPrompt, contextHistory, modelOptions) {
+      const initialPrompts = this.buildInitialPrompts(
+        systemPrompt,
+        contextHistory
+      );
+      const createOptions = {
+        ...modelOptions,
+        ...initialPrompts.length > 0 ? { initialPrompts } : {}
+      };
+      return api.create(createOptions);
+    }
+    getModelOptions() {
+      return {
         expectedInputs: [
           { type: "text", languages: this.expectedInputLanguages }
         ],
         expectedOutputs: [
           { type: "text", languages: this.expectedOutputLanguages }
         ]
-      });
+      };
+    }
+    buildInitialPrompts(systemPrompt, contextHistory) {
+      const configuredSystemPrompts = this.initialPrompts.filter((prompt) => prompt.role === "system").map((prompt) => prompt.content);
+      const combinedSystemPrompt = this.buildSystemPrompt(
+        [systemPrompt, ...configuredSystemPrompts].filter(Boolean).join("\n\n")
+      );
+      const prompts = [];
+      if (combinedSystemPrompt) {
+        prompts.push({ role: "system", content: combinedSystemPrompt });
+      }
+      prompts.push(
+        ...this.initialPrompts.filter((prompt) => prompt.role !== "system").map((prompt) => ({ ...prompt })),
+        ...contextHistory.filter(
+          (message) => message.role === "user" || message.role === "assistant"
+        ).map(({ role, content }) => ({ role, content }))
+      );
+      return prompts;
+    }
+    findLastUserMessageIndex(messages) {
+      for (let index = messages.length - 1; index >= 0; index -= 1) {
+        if (messages[index].role === "user") {
+          return index;
+        }
+      }
+      return -1;
     }
     buildSystemPrompt(systemPrompt) {
       const promptParts = [];
@@ -3680,7 +3879,20 @@ If it's in another language, summarize in that language.
       if (!maxTokens) {
         return void 0;
       }
-      return `Please keep your response concise, within approximately ${maxTokens} tokens.`;
+      switch (this._responseLength) {
+        case CHAT_RESPONSE_LENGTH.VERY_SHORT:
+          return `Reply in no more than one concise sentence. Do not use a preamble, summary, bullet points, or Markdown. Keep the reaction and answer short enough to read in one breath. Stay within approximately ${maxTokens} tokens.`;
+        case CHAT_RESPONSE_LENGTH.SHORT:
+          return `Reply in no more than two concise sentences. Do not use a preamble, summary, bullet points, or Markdown. Keep the reaction and answer short enough to read in one breath. Use natural conversational language. Stay within approximately ${maxTokens} tokens.`;
+        case CHAT_RESPONSE_LENGTH.MEDIUM:
+          return `Reply in no more than three concise sentences. Answer directly and include only the context needed to understand the answer. Do not use a preamble, summary, bullet points, or Markdown. Use natural conversational language. Stay within approximately ${maxTokens} tokens.`;
+        case CHAT_RESPONSE_LENGTH.LONG:
+          return `Reply in no more than five sentences. Explain the important details clearly, but avoid unnecessary preambles, repetition, bullet points, or Markdown. Use natural conversational language. Stay within approximately ${maxTokens} tokens.`;
+        case CHAT_RESPONSE_LENGTH.VERY_LONG:
+          return `Reply in no more than ten sentences. Give a thorough but focused answer. Use short plain-text paragraphs when helpful, and avoid unnecessary repetition, bullet points, or Markdown. Stay within approximately ${maxTokens} tokens.`;
+        case CHAT_RESPONSE_LENGTH.DEEP:
+          return `Give a detailed response and organize it in the way that best fits the request. No sentence-count limit applies at this level. Stay within approximately ${maxTokens} tokens.`;
+      }
     }
   };
 
@@ -3690,6 +3902,7 @@ If it's in another language, summarize in that language.
       return new GeminiNanoChatService({
         expectedInputLanguages: options.expectedInputLanguages,
         expectedOutputLanguages: options.expectedOutputLanguages,
+        initialPrompts: options.initialPrompts,
         responseLength: options.responseLength
       });
     }
@@ -3734,7 +3947,7 @@ If it's in another language, summarize in that language.
           `Model ${model} does not support reasoning_effort: ${reasoningEffort}.`
         );
       }
-      this.reasoningEffort = reasoningEffort ?? (isKimiReasoningEffortModel(model) ? supportedReasoningEfforts[0] : void 0);
+      this.reasoningEffort = reasoningEffort ?? getDefaultKimiReasoningEffort(model);
       this.visionModel = visionModel;
     }
     /**
@@ -3835,7 +4048,7 @@ If it's in another language, summarize in that language.
         body.response_format = this.responseFormat;
       }
       if (isKimiReasoningEffortModel(model)) {
-        body.reasoning_effort = this.reasoningEffort ?? "max";
+        body.reasoning_effort = this.reasoningEffort ?? getDefaultKimiReasoningEffort(model);
       }
       const effectiveThinking = this.resolveEffectiveThinking(model);
       if (effectiveThinking) {
@@ -4037,7 +4250,7 @@ If it's in another language, summarize in that language.
         }
         return void 0;
       }
-      const resolved = reasoningEffort ?? supported[0];
+      const resolved = reasoningEffort ?? getDefaultKimiReasoningEffort(model) ?? supported[0];
       if (!supported.includes(resolved)) {
         throw new Error(
           `Model ${model} supports reasoning_effort values: ${supported.join(", ")}.`
@@ -4391,8 +4604,19 @@ If it's in another language, summarize in that language.
   };
 
   // src/services/providers/openrouter/OpenRouterChatService.ts
+  var isOpenRouterAutoModel = (model) => [MODEL_OPENROUTER_AUTO, MODEL_OPENROUTER_AUTO_BETA].includes(model.trim());
   var isTokenLimitUnsupportedModel = (model) => [MODEL_GPT_OSS_20B_FREE, MODEL_ZAI_GLM_5_2].includes(model.trim());
   var getDefaultReasoningEffort = (model) => model.trim() === MODEL_ZAI_GLM_5_2 ? "none" : void 0;
+  var ensureAutoRouterOutput = (model, hasVisibleOutput) => {
+    if (isOpenRouterAutoModel(model) && !hasVisibleOutput) {
+      throw new Error(
+        "OpenRouter Auto Router returned no visible content. Please retry."
+      );
+    }
+  };
+  var hasUsableCompletionOutput = (completion) => completion.blocks.some(
+    (block) => block.type !== "text" || block.text.trim().length > 0
+  );
   var OpenRouterChatService = class {
     /**
      * Constructor
@@ -4476,7 +4700,7 @@ If it's in another language, summarize in that language.
         hasTools: this.tools.length > 0,
         runWithoutTools: async () => {
           const res = await this.callOpenRouter(messages, this.model, true);
-          return this.handleStream(res, onPartialResponse);
+          return this.handleStream(res, onPartialResponse, this.model);
         },
         runWithTools: () => this.chatOnce(messages, true, onPartialResponse),
         onCompleteResponse,
@@ -4505,7 +4729,7 @@ If it's in another language, summarize in that language.
               this.visionModel,
               true
             );
-            return this.handleStream(res, onPartialResponse);
+            return this.handleStream(res, onPartialResponse, this.visionModel);
           },
           runWithTools: () => this.visionChatOnce(messages, true, onPartialResponse),
           onCompleteResponse,
@@ -4533,7 +4757,7 @@ If it's in another language, summarize in that language.
         stream,
         maxTokens
       );
-      return stream ? this.parseStream(res, onPartialResponse) : this.parseOneShot(await res.json());
+      return stream ? this.parseStream(res, onPartialResponse, this.model) : this.parseOneShot(await res.json(), this.model);
     }
     /**
      * Process vision chat messages with tools
@@ -4557,7 +4781,7 @@ If it's in another language, summarize in that language.
         stream,
         maxTokens
       );
-      return stream ? this.parseStream(res, onPartialResponse) : this.parseOneShot(await res.json());
+      return stream ? this.parseStream(res, onPartialResponse, this.visionModel) : this.parseOneShot(await res.json(), this.visionModel);
     }
     /**
      * Call OpenRouter API
@@ -4586,7 +4810,8 @@ If it's in another language, summarize in that language.
         stream
       };
       const tokenLimit = maxTokens !== void 0 ? maxTokens : getMaxTokensForResponseLength(this.responseLength);
-      if (tokenLimit && isTokenLimitUnsupportedModel(model)) {
+      const shouldOmitTokenLimit = isTokenLimitUnsupportedModel(model) || isOpenRouterAutoModel(model) && maxTokens === void 0;
+      if (tokenLimit && shouldOmitTokenLimit) {
         console.warn(
           `OpenRouter: Token limits are disabled for ${model} because this model can return empty content when a token limit is set.`
         );
@@ -4621,24 +4846,34 @@ If it's in another language, summarize in that language.
      * Handle streaming response from OpenRouter
      * OpenRouter uses SSE format with potential comment lines
      */
-    async handleStream(res, onPartial) {
-      return parseOpenAICompatibleTextStream(res, onPartial, {
-        onJsonError: (payload) => console.debug("Failed to parse SSE data:", payload)
+    async handleStream(res, onPartial, model) {
+      const text = await parseOpenAICompatibleTextStream(res, onPartial, {
+        onJsonError: (payload) => console.debug("Failed to parse SSE data:", payload),
+        throwOnApiError: isOpenRouterAutoModel(model)
       });
+      ensureAutoRouterOutput(model, text.trim().length > 0);
+      return text;
     }
     /**
      * Parse streaming response with tool support
      */
-    async parseStream(res, onPartial) {
-      return parseOpenAICompatibleToolStream(res, onPartial, {
-        onJsonError: (payload) => console.debug("Failed to parse SSE data:", payload)
+    async parseStream(res, onPartial, model) {
+      const completion = await parseOpenAICompatibleToolStream(res, onPartial, {
+        onJsonError: (payload) => console.debug("Failed to parse SSE data:", payload),
+        throwOnApiError: isOpenRouterAutoModel(model)
       });
+      ensureAutoRouterOutput(model, hasUsableCompletionOutput(completion));
+      return completion;
     }
     /**
      * Parse non-streaming response
      */
-    parseOneShot(data) {
-      return parseOpenAICompatibleOneShot(data);
+    parseOneShot(data, model) {
+      const completion = parseOpenAICompatibleOneShot(data, {
+        throwOnApiError: isOpenRouterAutoModel(model)
+      });
+      ensureAutoRouterOutput(model, hasUsableCompletionOutput(completion));
+      return completion;
     }
   };
 
@@ -4691,6 +4926,7 @@ If it's in another language, summarize in that language.
       return [
         // OpenRouter routing
         MODEL_OPENROUTER_AUTO,
+        MODEL_OPENROUTER_AUTO_BETA,
         MODEL_OPENROUTER_FUSION,
         // Free models
         MODEL_GPT_OSS_20B_FREE,
@@ -4698,6 +4934,9 @@ If it's in another language, summarize in that language.
         // OpenAI models
         MODEL_OPENAI_GPT_LATEST,
         MODEL_OPENAI_GPT_MINI_LATEST,
+        MODEL_OPENAI_GPT_5_6_SOL,
+        MODEL_OPENAI_GPT_5_6_TERRA,
+        MODEL_OPENAI_GPT_5_6_LUNA,
         MODEL_OPENAI_GPT_5_5_PRO,
         MODEL_OPENAI_GPT_5_5,
         MODEL_OPENAI_GPT_5_1_CHAT,
@@ -4710,6 +4949,7 @@ If it's in another language, summarize in that language.
         // Anthropic models
         MODEL_ANTHROPIC_CLAUDE_SONNET_LATEST,
         MODEL_ANTHROPIC_CLAUDE_HAIKU_LATEST,
+        MODEL_ANTHROPIC_CLAUDE_OPUS_5,
         MODEL_ANTHROPIC_CLAUDE_OPUS_4,
         MODEL_ANTHROPIC_CLAUDE_SONNET_4,
         MODEL_ANTHROPIC_CLAUDE_3_7_SONNET,
@@ -4718,6 +4958,8 @@ If it's in another language, summarize in that language.
         // Gemini models
         MODEL_GOOGLE_GEMINI_PRO_LATEST,
         MODEL_GOOGLE_GEMINI_FLASH_LATEST,
+        MODEL_GOOGLE_GEMINI_3_6_FLASH,
+        MODEL_GOOGLE_GEMINI_3_5_FLASH_LITE,
         MODEL_GOOGLE_GEMINI_2_5_PRO,
         MODEL_GOOGLE_GEMINI_2_5_FLASH,
         MODEL_GOOGLE_GEMINI_2_5_FLASH_LITE_PREVIEW_09_2025,
@@ -4725,10 +4967,16 @@ If it's in another language, summarize in that language.
         MODEL_ZAI_GLM_5_2,
         MODEL_ZAI_GLM_4_7_FLASH,
         MODEL_ZAI_GLM_4_5_AIR,
+        // xAI models
+        MODEL_XAI_GROK_LATEST,
+        MODEL_XAI_GROK_4_5,
         // Other models
         MODEL_MOONSHOTAI_KIMI_LATEST,
+        MODEL_MOONSHOTAI_KIMI_K3,
         MODEL_MOONSHOTAI_KIMI_K2_7_CODE,
-        MODEL_MOONSHOTAI_KIMI_K2_5
+        MODEL_MOONSHOTAI_KIMI_K2_5,
+        MODEL_KWAIPILOT_KAT_CODER_AIR_V2_5,
+        MODEL_KWAIPILOT_KAT_CODER_PRO_V2_5
       ];
     }
     /**
@@ -5451,8 +5699,10 @@ If it's in another language, summarize in that language.
   ]);
   var REASONING_EFFORT_BY_PROVIDER = {
     openai: ["none", "minimal", "low", "medium", "high", "xhigh", "max"],
+    claude: ["low", "medium", "high", "xhigh", "max"],
     gemini: ["minimal", "low", "medium", "high"],
     openrouter: ["none", "minimal", "low", "medium", "high"],
+    kimi: ["low", "high", "max"],
     mistral: ["low", "medium", "high"],
     plamo: ["none", "medium"],
     xai: ["none", "low", "medium", "high"]
@@ -5515,7 +5765,7 @@ If it's in another language, summarize in that language.
         mcp: MCP_SUPPORTED_PROVIDERS.has(providerName),
         jsonMode: JSON_MODE_SUPPORTED_PROVIDERS.has(providerName),
         responseLength: true,
-        reasoningEffort: REASONING_EFFORT_BY_PROVIDER[providerName] ?? []
+        reasoningEffort: providerName === "claude" && model ? [...getClaudeSupportedReasoningEfforts(model)] : providerName === "kimi" && model ? [...getKimiSupportedReasoningEfforts(model)] : REASONING_EFFORT_BY_PROVIDER[providerName] ?? []
       };
     }
     /**