@deepstrike/sdk 0.2.10 → 0.2.12

package/dist/providers/gemini.js

@@ -1,6 +1,6 @@
 import { GoogleGenerativeAI } from "@google/generative-ai";
 import { withServerRuntimeGuard } from "../runtime/server.js";
-import { CircuitBreaker, normalizeToolCall } from "./base.js";
+import { CircuitBreaker, normalizeToolCall, turnsWithStateAppended } from "./base.js";
 import { endpointProfiles } from "./profiles.js";
 const GEMINI_BASE = endpointProfiles["gemini.google"].baseURL;
 const GEMINI_POLICIES = {
@@ -56,8 +56,23 @@ export function buildContents(turns) {
                 parts.push({ functionCall: { name: tc.name, args } });
             }
         }
-        if (msg.content)
+        // Multimodal: render contentParts (text + image) when present, else the plain
+        // text body. Without this, image inputs to Gemini were silently dropped.
+        if (msg.contentParts?.length) {
+            for (const p of msg.contentParts) {
+                if (p.type === "text")
+                    parts.push({ text: p.text });
+                else if (p.type === "image") {
+                    if (p.data)
+                        parts.push({ inlineData: { mimeType: p.mediaType ?? "image/png", data: p.data } });
+                    else if (p.url)
+                        parts.push({ fileData: { mimeType: p.mediaType ?? "image/png", fileUri: p.url } });
+                }
+            }
+        }
+        else if (msg.content) {
             parts.push({ text: msg.content });
+        }
         if (parts.length)
             contents.push({ role, parts });
     }
@@ -96,7 +111,7 @@ export class GeminiProvider {
         if (this.circuit.isOpen())
             throw new Error("Circuit breaker open");
         const system = context.systemText || undefined;
-        const contents = buildContents(context.turns);
+        const contents = buildContents(turnsWithStateAppended(context));
         const geminiTools = buildTools(tools);
         let lastErr;
         for (let i = 0; i < this.maxRetries; i++) {
@@ -140,7 +155,7 @@ export class GeminiProvider {
     }
     async *stream(context, tools, extensions) {
         const system = context.systemText || undefined;
-        const contents = buildContents(context.turns);
+        const contents = buildContents(turnsWithStateAppended(context));
         const geminiTools = buildTools(tools);
         const m = this.genAI.getGenerativeModel({
             ...this.modelExtensions(extensions),
@@ -165,11 +180,15 @@ export class GeminiProvider {
         }
         const usage = (await result.response).usageMetadata;
         if (usage?.totalTokenCount) {
+            // Gemini implicit/explicit cache hits are reported as cachedContentTokenCount,
+            // a subset of promptTokenCount (which stays the full prompt for accounting).
+            const cachedTokens = usage.cachedContentTokenCount ?? 0;
             yield {
                 type: "usage",
                 totalTokens: usage.totalTokenCount,
                 inputTokens: usage.promptTokenCount ?? 0,
                 outputTokens: usage.candidatesTokenCount ?? 0,
+                ...(cachedTokens > 0 ? { cacheReadInputTokens: cachedTokens } : {}),
             };
         }
     }

package/dist/providers/glm.d.ts

@@ -1,5 +1,17 @@
 import type { ProviderDescriptor, RuntimePolicy } from "../types.js";
+import { AnthropicProvider } from "./anthropic.js";
 import { OpenAIChatProvider } from "./openai.js";
+/**
+ * GLM over its Anthropic-compatible endpoint.
+ */
+export declare class GLMAnthropicProvider extends AnthropicProvider {
+    constructor(apiKey: string, model?: string, retry?: {
+        maxRetries: number;
+        baseDelay: number;
+    }, baseURL?: string);
+    protected providerName(): string;
+    runtimePolicy(): RuntimePolicy;
+}
 export declare class GLMProvider extends OpenAIChatProvider {
     constructor(apiKey: string, model?: string, retry?: {
         maxRetries: number;

package/dist/providers/glm.js

@@ -1,6 +1,6 @@
+import { AnthropicProvider } from "./anthropic.js";
 import { OpenAIChatProvider } from "./openai.js";
 import { endpointProfiles } from "./profiles.js";
-const GLM_BASE = endpointProfiles["glm.openai"].baseURL;
 const GLM_POLICIES = {
     "glm-5.1": { maxTurns: 50 },
     "glm/glm-5.1": { maxTurns: 50 },
@@ -11,8 +11,25 @@ const GLM_POLICIES = {
     "glm-4-air": { maxTurns: 20 },
     "glm/glm-4-air": { maxTurns: 20 },
 };
+/**
+ * GLM over its Anthropic-compatible endpoint.
+ */
+export class GLMAnthropicProvider extends AnthropicProvider {
+    constructor(apiKey, model = "glm-5.1", retry, baseURL = endpointProfiles["glm.anthropic"].baseURL) {
+        super(apiKey, model, retry, {
+            baseURL,
+            authMode: "api-key",
+        });
+    }
+    providerName() {
+        return "glm";
+    }
+    runtimePolicy() {
+        return GLM_POLICIES[this.model] ?? {};
+    }
+}
 export class GLMProvider extends OpenAIChatProvider {
-    constructor(apiKey, model = "glm-5.1", retry, baseURL = GLM_BASE) {
+    constructor(apiKey, model = "glm-5.1", retry, baseURL = endpointProfiles["glm.openai"].baseURL) {
         super(apiKey, model, retry, baseURL);
     }
     runtimePolicy() {

package/dist/providers/kimi.d.ts

@@ -1,5 +1,17 @@
 import type { ProviderDescriptor, RuntimePolicy } from "../types.js";
+import { AnthropicProvider } from "./anthropic.js";
 import { OpenAIChatProvider } from "./openai.js";
+/**
+ * Kimi over its Anthropic-compatible endpoint.
+ */
+export declare class KimiAnthropicProvider extends AnthropicProvider {
+    constructor(apiKey: string, model?: string, retry?: {
+        maxRetries: number;
+        baseDelay: number;
+    }, baseURL?: string);
+    protected providerName(): string;
+    runtimePolicy(): RuntimePolicy;
+}
 export declare class KimiProvider extends OpenAIChatProvider {
     constructor(apiKey: string, model?: string, retry?: {
         maxRetries: number;

package/dist/providers/kimi.js

@@ -1,6 +1,6 @@
+import { AnthropicProvider } from "./anthropic.js";
 import { OpenAIChatProvider } from "./openai.js";
 import { endpointProfiles } from "./profiles.js";
-const MOONSHOT_BASE = endpointProfiles["kimi.openai"].baseURL;
 const KIMI_POLICIES = {
     "moonshot-v1-8k": { maxTurns: 15 },
     "moonshot-v1-32k": { maxTurns: 20 },
@@ -10,8 +10,25 @@ const KIMI_POLICIES = {
     "kimi-k2-thinking": { maxTurns: 50 },
     "kimi-k2-thinking-turbo": { maxTurns: 40 },
 };
+/**
+ * Kimi over its Anthropic-compatible endpoint.
+ */
+export class KimiAnthropicProvider extends AnthropicProvider {
+    constructor(apiKey, model = "kimi-k2.6", retry, baseURL = endpointProfiles["kimi.anthropic"].baseURL) {
+        super(apiKey, model, retry, {
+            baseURL,
+            authMode: "api-key",
+        });
+    }
+    providerName() {
+        return "kimi";
+    }
+    runtimePolicy() {
+        return KIMI_POLICIES[this.model] ?? {};
+    }
+}
 export class KimiProvider extends OpenAIChatProvider {
-    constructor(apiKey, model = "kimi-k2.6", retry, baseURL = MOONSHOT_BASE) {
+    constructor(apiKey, model = "kimi-k2.6", retry, baseURL = endpointProfiles["kimi.openai"].baseURL) {
         super(apiKey, model, retry, baseURL);
     }
     runtimePolicy() {

package/dist/providers/minimax.js

@@ -1,7 +1,7 @@
 import { AnthropicProvider } from "./anthropic.js";
 import { OpenAIChatProvider } from "./openai.js";
 import { endpointProfiles } from "./profiles.js";
-import { omitExtensionKeys } from "./base.js";
+import { omitExtensionKeys, openAICachedPromptTokens } from "./base.js";
 const MINIMAX_POLICIES = {
     "MiniMax-M2.7": { maxTurns: 35 },
     "MiniMax-M2.7-highspeed": { maxTurns: 35 },
@@ -127,11 +127,13 @@ export class MiniMaxOpenAIProvider extends OpenAIChatProvider {
         let totalTokens = 0;
         let inputTokens = 0;
         let outputTokens = 0;
+        let cacheReadTokens = 0;
         for await (const chunk of stream) {
             if (chunk.usage) {
                 totalTokens = chunk.usage.total_tokens;
                 inputTokens = chunk.usage.prompt_tokens ?? 0;
                 outputTokens = chunk.usage.completion_tokens ?? 0;
+                cacheReadTokens = openAICachedPromptTokens(chunk.usage);
                 continue;
             }
             const choice = chunk.choices[0];
@@ -195,7 +197,7 @@ export class MiniMaxOpenAIProvider extends OpenAIChatProvider {
             yield { type: "tool_call", id: tb.id, name: tb.name, arguments: args };
         }
         if (totalTokens > 0)
-            yield { type: "usage", totalTokens, inputTokens, outputTokens };
+            yield { type: "usage", totalTokens, inputTokens, outputTokens, ...(cacheReadTokens > 0 ? { cacheReadInputTokens: cacheReadTokens } : {}) };
     }
     rememberMiniMaxReplay(content, toolCalls, reasoningContent, reasoningDetails, nativeToolCalls) {
         const hasReasoning = typeof reasoningContent === "string" && reasoningContent.trim().length > 0;

package/dist/providers/ollama.js

@@ -1,4 +1,4 @@
-import { normalizeToolCall, omitExtensionKeys } from "./base.js";
+import { normalizeToolCall, omitExtensionKeys, turnsWithStateAppended } from "./base.js";
 // Prefix-based policy for local models (first match wins)
 const OLLAMA_PREFIX_POLICIES = [
     ["deepseek-r1", { maxTurns: 40 }],
@@ -32,7 +32,7 @@ export class OllamaProvider {
         const result = [];
         if (context.systemText)
             result.push({ role: "system", content: context.systemText });
-        for (const m of context.turns) {
+        for (const m of turnsWithStateAppended(context)) {
             const images = [];
             if (m.contentParts?.length) {
                 for (const p of m.contentParts) {

package/dist/providers/openai-chat.js

@@ -17,9 +17,10 @@ export class OpenAIChatAdapter {
             replayForAssistant: message => this.replayFields.get(assistantReplayKey(message)),
         });
         const degradeReasoning = Boolean(options.requireNonEmptyReasoningForToolCalls && options.degradeMissingReasoning);
-        // toOpenAIMessageParams prepends systemText as messages[0], then turns.
+        // toOpenAIMessageParams emits [system?, ...history, stateTurn?] — the State
+        // turn is APPENDED last, so the cursor only skips the system message to align
+        // with context.turns (history) when writing reasoning-replay fields.
         const serialized = toOpenAIMessageParams(context);
-        // Cursor starts at 1 to skip the system message injected by toOpenAIMessageParams.
         let cursor = context.systemText ? 1 : 0;
         for (const source of context.turns) {
             if (source.role === "tool") {

package/dist/providers/openai-responses.js

@@ -55,6 +55,15 @@ export class OpenAIResponsesAdapter {
                 content: this.buildMessageContent(message),
             });
         }
+        // The volatile State turn is sent every turn (it changes each call and is
+        // never "covered" by previous_response_id). Absent on un-rebuilt bindings,
+        // where the state is already inside the covered/uncovered history.
+        if (context.stateTurn) {
+            input.push({
+                role: context.stateTurn.role === "assistant" ? "assistant" : "user",
+                content: this.buildMessageContent(context.stateTurn),
+            });
+        }
         return input;
     }
     decodeOutput(output) {
@@ -223,11 +232,15 @@ export class OpenAIResponsesProvider {
                 runState.previousResponseId = evt.response.id;
                 runState.coveredMessageCount = context.turns.length + 1;
                 if (evt.response.usage?.total_tokens) {
+                    // Responses API reports prompt-cache hits as input_tokens_details.cached_tokens,
+                    // a subset of input_tokens (the full prompt, kept for accounting).
+                    const cachedTokens = evt.response.usage.input_tokens_details?.cached_tokens ?? 0;
                     yield {
                         type: "usage",
                         totalTokens: evt.response.usage.total_tokens,
                         ...(evt.response.usage.input_tokens ? { inputTokens: evt.response.usage.input_tokens } : {}),
                         ...(evt.response.usage.output_tokens ? { outputTokens: evt.response.usage.output_tokens } : {}),
+                        ...(cachedTokens > 0 ? { cacheReadInputTokens: cachedTokens } : {}),
                     };
                 }
             }

package/dist/providers/openai.d.ts

@@ -31,6 +31,13 @@ export declare class OpenAIChatProvider implements LLMProvider {
     seedProviderReplay(message: Pick<Message, "content" | "toolCalls">, replay: ProviderReplay): void;
     complete(context: RenderedContext, tools: ToolSchema[], extensions?: Record<string, unknown>): Promise<Message>;
     stream(context: RenderedContext, tools: ToolSchema[], extensions?: Record<string, unknown>): AsyncIterable<StreamEvent>;
+    /**
+     * Default `prompt_cache_key` derived from the cacheable prefix (system prompt +
+     * tool names) so requests for the same agent config route to the same cache.
+     * A caller-supplied `prompt_cache_key` in extensions overrides it (it is spread
+     * after this default). Unknown to non-OpenAI compatible endpoints, which ignore it.
+     */
+    protected promptCacheKey(context: RenderedContext, tools: ToolSchema[]): string;
     protected requestExtensions(extensions?: Record<string, unknown>): Record<string, unknown>;
 }
 export { OpenAIChatProvider as OpenAIProvider };

package/dist/providers/openai.js

@@ -1,6 +1,6 @@
 import OpenAI from "openai";
 import { withServerRuntimeGuard } from "../runtime/server.js";
-import { CircuitBreaker, omitExtensionKeys, ThinkingTagStreamExtractor } from "./base.js";
+import { CircuitBreaker, omitExtensionKeys, openAICachedPromptTokens, stablePromptCacheKey, ThinkingTagStreamExtractor } from "./base.js";
 import { OpenAIChatAdapter } from "./openai-chat.js";
 const OPENAI_POLICIES = {
     "gpt-5.5": { maxTurns: 60 },
@@ -102,6 +102,7 @@ export class OpenAIChatProvider {
         for (let i = 0; i < this.maxRetries; i++) {
             try {
                 const resp = await this.client.chat.completions.create({
+                    prompt_cache_key: this.promptCacheKey(context, tools),
                     ...this.requestExtensions(extensions),
                     model: this.model,
                     messages: msgs,
@@ -129,6 +130,7 @@ export class OpenAIChatProvider {
         let accumulatedReasoning = "";
         let accumulatedContent = "";
         const stream = await this.client.chat.completions.create({
+            prompt_cache_key: this.promptCacheKey(context, tools),
             ...this.requestExtensions(extensions),
             model: this.model,
             messages: msgs,
@@ -139,11 +141,13 @@ export class OpenAIChatProvider {
         let totalTokens = 0;
         let inputTokens = 0;
         let outputTokens = 0;
+        let cacheReadTokens = 0;
         for await (const chunk of stream) {
             if (chunk.usage) {
                 totalTokens = chunk.usage.total_tokens;
                 inputTokens = chunk.usage.prompt_tokens ?? 0;
                 outputTokens = chunk.usage.completion_tokens ?? 0;
+                cacheReadTokens = openAICachedPromptTokens(chunk.usage);
                 continue;
             }
             const choice = chunk.choices[0];
@@ -228,7 +232,16 @@ export class OpenAIChatProvider {
             yield { type: "tool_call", id: tb.id, name: tb.name, arguments: args };
         }
         if (totalTokens > 0)
-            yield { type: "usage", totalTokens, inputTokens, outputTokens };
+            yield { type: "usage", totalTokens, inputTokens, outputTokens, ...(cacheReadTokens > 0 ? { cacheReadInputTokens: cacheReadTokens } : {}) };
+    }
+    /**
+     * Default `prompt_cache_key` derived from the cacheable prefix (system prompt +
+     * tool names) so requests for the same agent config route to the same cache.
+     * A caller-supplied `prompt_cache_key` in extensions overrides it (it is spread
+     * after this default). Unknown to non-OpenAI compatible endpoints, which ignore it.
+     */
+    promptCacheKey(context, tools) {
+        return stablePromptCacheKey([context.systemText, tools.map(t => t.name).join(",")]);
     }
     requestExtensions(extensions) {
         return omitExtensionKeys(extensions, ["model", "messages", "tools", "stream", "stream_options", "__deepstrikeThinkingEnabled"]);

package/dist/providers/profiles.d.ts

@@ -65,18 +65,36 @@ export declare const endpointProfiles: {
         readonly protocol: "openai-chat";
         readonly baseURL: "https://api.minimaxi.com/v1";
     };
+    readonly "deepseek.anthropic": {
+        readonly id: "deepseek.anthropic";
+        readonly providerId: "deepseek";
+        readonly protocol: "anthropic-messages";
+        readonly baseURL: "https://api.deepseek.com/anthropic";
+    };
     readonly "deepseek.openai": {
         readonly id: "deepseek.openai";
         readonly providerId: "deepseek";
         readonly protocol: "openai-chat";
         readonly baseURL: "https://api.deepseek.com";
     };
+    readonly "kimi.anthropic": {
+        readonly id: "kimi.anthropic";
+        readonly providerId: "kimi";
+        readonly protocol: "anthropic-messages";
+        readonly baseURL: "https://api.moonshot.ai/anthropic";
+    };
     readonly "kimi.openai": {
         readonly id: "kimi.openai";
         readonly providerId: "kimi";
         readonly protocol: "openai-chat";
         readonly baseURL: "https://api.moonshot.cn/v1";
     };
+    readonly "qwen.anthropic": {
+        readonly id: "qwen.anthropic";
+        readonly providerId: "qwen";
+        readonly protocol: "anthropic-messages";
+        readonly baseURL: "https://dashscope-intl.aliyuncs.com/apps/anthropic";
+    };
     readonly "qwen.dashscope": {
         readonly id: "qwen.dashscope";
         readonly providerId: "qwen";
@@ -107,6 +125,12 @@ export declare const endpointProfiles: {
         readonly protocol: "gemini-embeddings";
         readonly baseURL: "https://generativelanguage.googleapis.com";
     };
+    readonly "glm.anthropic": {
+        readonly id: "glm.anthropic";
+        readonly providerId: "glm";
+        readonly protocol: "anthropic-messages";
+        readonly baseURL: "https://api.z.ai/api/anthropic";
+    };
     readonly "glm.openai": {
         readonly id: "glm.openai";
         readonly providerId: "glm";
@@ -868,7 +892,7 @@ export declare const modelProfiles: {
     readonly "deepseek/deepseek-chat": {
         readonly id: "deepseek/deepseek-chat";
         readonly providerId: "deepseek";
-        readonly defaultEndpointId: "deepseek.openai";
+        readonly defaultEndpointId: "deepseek.anthropic";
         readonly contextWindow: 64000;
         readonly modalities: {
             readonly input: ["text"];
@@ -888,7 +912,7 @@ export declare const modelProfiles: {
     readonly "deepseek/deepseek-reasoner": {
         readonly id: "deepseek/deepseek-reasoner";
         readonly providerId: "deepseek";
-        readonly defaultEndpointId: "deepseek.openai";
+        readonly defaultEndpointId: "deepseek.anthropic";
         readonly contextWindow: 64000;
         readonly modalities: {
             readonly input: ["text"];
@@ -908,7 +932,7 @@ export declare const modelProfiles: {
     readonly "deepseek/deepseek-v4-flash": {
         readonly id: "deepseek/deepseek-v4-flash";
         readonly providerId: "deepseek";
-        readonly defaultEndpointId: "deepseek.openai";
+        readonly defaultEndpointId: "deepseek.anthropic";
         readonly contextWindow: 1000000;
         readonly modalities: {
             readonly input: ["text"];
@@ -928,7 +952,7 @@ export declare const modelProfiles: {
     readonly "deepseek/deepseek-v4-pro": {
         readonly id: "deepseek/deepseek-v4-pro";
         readonly providerId: "deepseek";
-        readonly defaultEndpointId: "deepseek.openai";
+        readonly defaultEndpointId: "deepseek.anthropic";
         readonly contextWindow: 1000000;
         readonly modalities: {
             readonly input: ["text"];
@@ -948,7 +972,7 @@ export declare const modelProfiles: {
     readonly "kimi/moonshot-v1-8k": {
         readonly id: "kimi/moonshot-v1-8k";
         readonly providerId: "kimi";
-        readonly defaultEndpointId: "kimi.openai";
+        readonly defaultEndpointId: "kimi.anthropic";
         readonly contextWindow: 8000;
         readonly modalities: {
             readonly input: ["text"];
@@ -968,7 +992,7 @@ export declare const modelProfiles: {
     readonly "kimi/moonshot-v1-32k": {
         readonly id: "kimi/moonshot-v1-32k";
         readonly providerId: "kimi";
-        readonly defaultEndpointId: "kimi.openai";
+        readonly defaultEndpointId: "kimi.anthropic";
         readonly contextWindow: 32000;
         readonly modalities: {
             readonly input: ["text"];
@@ -988,7 +1012,7 @@ export declare const modelProfiles: {
     readonly "kimi/moonshot-v1-128k": {
         readonly id: "kimi/moonshot-v1-128k";
         readonly providerId: "kimi";
-        readonly defaultEndpointId: "kimi.openai";
+        readonly defaultEndpointId: "kimi.anthropic";
         readonly contextWindow: 128000;
         readonly modalities: {
             readonly input: ["text", "image"];
@@ -1008,7 +1032,7 @@ export declare const modelProfiles: {
     readonly "kimi/kimi-k2.5": {
         readonly id: "kimi/kimi-k2.5";
         readonly providerId: "kimi";
-        readonly defaultEndpointId: "kimi.openai";
+        readonly defaultEndpointId: "kimi.anthropic";
         readonly contextWindow: 256000;
         readonly modalities: {
             readonly input: ["text", "image"];
@@ -1028,7 +1052,7 @@ export declare const modelProfiles: {
     readonly "kimi/kimi-k2.6": {
         readonly id: "kimi/kimi-k2.6";
         readonly providerId: "kimi";
-        readonly defaultEndpointId: "kimi.openai";
+        readonly defaultEndpointId: "kimi.anthropic";
         readonly modalities: {
             readonly input: ["text"];
             readonly output: ["text"];
@@ -1047,7 +1071,7 @@ export declare const modelProfiles: {
     readonly "kimi/kimi-k2-thinking": {
         readonly id: "kimi/kimi-k2-thinking";
         readonly providerId: "kimi";
-        readonly defaultEndpointId: "kimi.openai";
+        readonly defaultEndpointId: "kimi.anthropic";
         readonly contextWindow: 256000;
         readonly modalities: {
             readonly input: ["text"];
@@ -1067,7 +1091,7 @@ export declare const modelProfiles: {
     readonly "kimi/kimi-k2-thinking-turbo": {
         readonly id: "kimi/kimi-k2-thinking-turbo";
         readonly providerId: "kimi";
-        readonly defaultEndpointId: "kimi.openai";
+        readonly defaultEndpointId: "kimi.anthropic";
         readonly contextWindow: 256000;
         readonly modalities: {
             readonly input: ["text"];
@@ -1087,7 +1111,7 @@ export declare const modelProfiles: {
     readonly "qwen/qwen3.7-max-preview": {
         readonly id: "qwen/qwen3.7-max-preview";
         readonly providerId: "qwen";
-        readonly defaultEndpointId: "qwen.dashscope";
+        readonly defaultEndpointId: "qwen.anthropic";
         readonly contextWindow: 256000;
         readonly modalities: {
             readonly input: ["text"];
@@ -1107,7 +1131,7 @@ export declare const modelProfiles: {
     readonly "qwen/qwen3.7-plus-preview": {
         readonly id: "qwen/qwen3.7-plus-preview";
         readonly providerId: "qwen";
-        readonly defaultEndpointId: "qwen.dashscope";
+        readonly defaultEndpointId: "qwen.anthropic";
         readonly contextWindow: 1000000;
         readonly modalities: {
             readonly input: ["text", "image"];
@@ -1127,7 +1151,7 @@ export declare const modelProfiles: {
     readonly "qwen/qwen3.6-max-preview": {
         readonly id: "qwen/qwen3.6-max-preview";
         readonly providerId: "qwen";
-        readonly defaultEndpointId: "qwen.dashscope";
+        readonly defaultEndpointId: "qwen.anthropic";
         readonly contextWindow: 256000;
         readonly modalities: {
             readonly input: ["text"];
@@ -1147,7 +1171,7 @@ export declare const modelProfiles: {
     readonly "qwen/qwen3.6-plus": {
         readonly id: "qwen/qwen3.6-plus";
         readonly providerId: "qwen";
-        readonly defaultEndpointId: "qwen.dashscope";
+        readonly defaultEndpointId: "qwen.anthropic";
         readonly contextWindow: 1000000;
         readonly modalities: {
             readonly input: ["text", "image"];
@@ -1167,7 +1191,7 @@ export declare const modelProfiles: {
     readonly "qwen/qwen3.6-flash": {
         readonly id: "qwen/qwen3.6-flash";
         readonly providerId: "qwen";
-        readonly defaultEndpointId: "qwen.dashscope";
+        readonly defaultEndpointId: "qwen.anthropic";
         readonly contextWindow: 1000000;
         readonly modalities: {
             readonly input: ["text", "image"];
@@ -1187,7 +1211,7 @@ export declare const modelProfiles: {
     readonly "qwen/qwen3.6-35b-a3b": {
         readonly id: "qwen/qwen3.6-35b-a3b";
         readonly providerId: "qwen";
-        readonly defaultEndpointId: "qwen.dashscope";
+        readonly defaultEndpointId: "qwen.anthropic";
         readonly contextWindow: 256000;
         readonly modalities: {
             readonly input: ["text", "image"];
@@ -1207,7 +1231,7 @@ export declare const modelProfiles: {
     readonly "qwen/qwen3.6-27b": {
         readonly id: "qwen/qwen3.6-27b";
         readonly providerId: "qwen";
-        readonly defaultEndpointId: "qwen.dashscope";
+        readonly defaultEndpointId: "qwen.anthropic";
         readonly contextWindow: 256000;
         readonly modalities: {
             readonly input: ["text", "image"];
@@ -1227,7 +1251,7 @@ export declare const modelProfiles: {
     readonly "qwen/qwen3.5-plus": {
         readonly id: "qwen/qwen3.5-plus";
         readonly providerId: "qwen";
-        readonly defaultEndpointId: "qwen.dashscope";
+        readonly defaultEndpointId: "qwen.anthropic";
         readonly contextWindow: 1000000;
         readonly modalities: {
             readonly input: ["text", "image"];
@@ -1247,7 +1271,7 @@ export declare const modelProfiles: {
     readonly "qwen/qwen3.5-flash": {
         readonly id: "qwen/qwen3.5-flash";
         readonly providerId: "qwen";
-        readonly defaultEndpointId: "qwen.dashscope";
+        readonly defaultEndpointId: "qwen.anthropic";
         readonly contextWindow: 1000000;
         readonly modalities: {
             readonly input: ["text", "image"];
@@ -1267,7 +1291,7 @@ export declare const modelProfiles: {
     readonly "qwen/qwen3.5-397b-a17b": {
         readonly id: "qwen/qwen3.5-397b-a17b";
         readonly providerId: "qwen";
-        readonly defaultEndpointId: "qwen.dashscope";
+        readonly defaultEndpointId: "qwen.anthropic";
         readonly contextWindow: 256000;
         readonly modalities: {
             readonly input: ["text", "image"];
@@ -1287,7 +1311,7 @@ export declare const modelProfiles: {
     readonly "qwen/qwen3.5-122b-a10b": {
         readonly id: "qwen/qwen3.5-122b-a10b";
         readonly providerId: "qwen";
-        readonly defaultEndpointId: "qwen.dashscope";
+        readonly defaultEndpointId: "qwen.anthropic";
         readonly contextWindow: 256000;
         readonly modalities: {
             readonly input: ["text", "image"];
@@ -1307,7 +1331,7 @@ export declare const modelProfiles: {
     readonly "qwen/qwen3.5-35b-a3b": {
         readonly id: "qwen/qwen3.5-35b-a3b";
         readonly providerId: "qwen";
-        readonly defaultEndpointId: "qwen.dashscope";
+        readonly defaultEndpointId: "qwen.anthropic";
         readonly contextWindow: 256000;
         readonly modalities: {
             readonly input: ["text", "image"];
@@ -1327,7 +1351,7 @@ export declare const modelProfiles: {
     readonly "qwen/qwen3.5-27b": {
         readonly id: "qwen/qwen3.5-27b";
         readonly providerId: "qwen";
-        readonly defaultEndpointId: "qwen.dashscope";
+        readonly defaultEndpointId: "qwen.anthropic";
         readonly contextWindow: 256000;
         readonly modalities: {
             readonly input: ["text", "image"];
@@ -1629,7 +1653,7 @@ export declare const modelProfiles: {
     readonly "glm/glm-5.1": {
         readonly id: "glm/glm-5.1";
         readonly providerId: "glm";
-        readonly defaultEndpointId: "glm.openai";
+        readonly defaultEndpointId: "glm.anthropic";
         readonly contextWindow: 200000;
         readonly modalities: {
             readonly input: ["text"];
@@ -1649,7 +1673,7 @@ export declare const modelProfiles: {
     readonly "glm/glm-4-plus": {
         readonly id: "glm/glm-4-plus";
         readonly providerId: "glm";
-        readonly defaultEndpointId: "glm.openai";
+        readonly defaultEndpointId: "glm.anthropic";
         readonly contextWindow: 128000;
         readonly modalities: {
             readonly input: ["text", "image"];
@@ -1669,7 +1693,7 @@ export declare const modelProfiles: {
     readonly "glm/glm-4-flash": {
         readonly id: "glm/glm-4-flash";
         readonly providerId: "glm";
-        readonly defaultEndpointId: "glm.openai";
+        readonly defaultEndpointId: "glm.anthropic";
         readonly contextWindow: 128000;
         readonly modalities: {
             readonly input: ["text"];
@@ -1689,7 +1713,7 @@ export declare const modelProfiles: {
     readonly "glm/glm-4-air": {
         readonly id: "glm/glm-4-air";
         readonly providerId: "glm";
-        readonly defaultEndpointId: "glm.openai";
+        readonly defaultEndpointId: "glm.anthropic";
         readonly contextWindow: 128000;
         readonly modalities: {
             readonly input: ["text"];