@llmgateway/models 1.326.0 → 1.327.0

package/dist/models/alibaba.d.ts

@@ -1661,6 +1661,7 @@ export declare const alibabaModels: [{
     }, {
         readonly providerId: "deepinfra";
         readonly externalId: "Qwen/Qwen3.8-27B";
+        readonly deactivatedAt: Date;
         readonly inputPrice: "0.2e-6";
         readonly cachedInputPrice: "0.05e-6";
         readonly outputPrice: "2.5e-6";
@@ -1712,6 +1713,7 @@ export declare const alibabaModels: [{
         readonly maxOutput: 131072;
         readonly quantization: "fp4";
         readonly maxTemperature: 1;
+        readonly minCacheableTokens: 4096;
         readonly streaming: true;
         readonly reasoning: true;
         readonly reasoningEfforts: ["minimal", "low", "medium", "high", "xhigh", "max"];
@@ -3019,6 +3021,104 @@ export declare const alibabaModels: [{
         readonly jsonOutput: true;
         readonly supportedParameters: ["temperature", "max_tokens", "top_p", "frequency_penalty", "presence_penalty", "stop", "stream", "response_format", "tools", "reasoning_effort"];
     }];
+}, {
+    readonly id: "qwen3.5-122b-a10b";
+    readonly name: "Qwen3.5 122B A10B";
+    readonly description: "Open-weight Qwen3.5 native vision-language MoE model (122B total, 10B active) with hybrid linear attention.";
+    readonly family: "alibaba";
+    readonly releasedAt: Date;
+    readonly providers: [{
+        readonly providerId: "alibaba";
+        readonly externalId: "qwen3.5-122b-a10b";
+        readonly inputPrice: "0.4e-6";
+        readonly outputPrice: "3.2e-6";
+        readonly regions: [{
+            readonly id: "singapore";
+        }];
+        readonly requestPrice: "0";
+        readonly contextSize: 262144;
+        readonly maxOutput: 65536;
+        readonly maxTemperature: 1;
+        readonly reasoning: true;
+        readonly reasoningEfforts: ["none", "minimal", "low", "medium", "high", "xhigh", "max"];
+        readonly reasoningMaxTokens: true;
+        readonly streaming: true;
+        readonly vision: true;
+        readonly tools: true;
+        readonly jsonOutput: true;
+        readonly supportedParameters: ["temperature", "max_tokens", "top_p", "frequency_penalty", "presence_penalty", "stop", "stream", "response_format", "tools", "reasoning_effort"];
+    }, {
+        readonly providerId: "novita";
+        readonly externalId: "qwen/qwen3.5-122b-a10b";
+        readonly inputPrice: "0.4e-6";
+        readonly outputPrice: "3.2e-6";
+        readonly requestPrice: "0";
+        readonly contextSize: 262144;
+        readonly maxOutput: 65536;
+        readonly quantization: "bf16";
+        readonly maxTemperature: 1;
+        readonly reasoning: true;
+        readonly streaming: true;
+        readonly vision: true;
+        readonly requiresBase64Images: true;
+        readonly tools: true;
+        readonly jsonOutput: true;
+        readonly supportedParameters: ["temperature", "max_tokens", "top_p", "frequency_penalty", "presence_penalty", "stop", "stream", "response_format", "tools"];
+    }];
+}, {
+    readonly id: "qwen3.5-27b";
+    readonly name: "Qwen3.5 27B";
+    readonly description: "Open-weight dense Qwen3.5 native vision-language model with linear attention for fast responses.";
+    readonly family: "alibaba";
+    readonly releasedAt: Date;
+    readonly providers: [{
+        readonly providerId: "alibaba";
+        readonly externalId: "qwen3.5-27b";
+        readonly inputPrice: "0.3e-6";
+        readonly outputPrice: "2.4e-6";
+        readonly regions: [{
+            readonly id: "singapore";
+        }];
+        readonly requestPrice: "0";
+        readonly contextSize: 262144;
+        readonly maxOutput: 65536;
+        readonly maxTemperature: 1;
+        readonly reasoning: true;
+        readonly reasoningEfforts: ["none", "minimal", "low", "medium", "high", "xhigh", "max"];
+        readonly reasoningMaxTokens: true;
+        readonly streaming: true;
+        readonly vision: true;
+        readonly tools: true;
+        readonly jsonOutput: true;
+        readonly supportedParameters: ["temperature", "max_tokens", "top_p", "frequency_penalty", "presence_penalty", "stop", "stream", "response_format", "tools", "reasoning_effort"];
+    }];
+}, {
+    readonly id: "qwen3.5-35b-a3b";
+    readonly name: "Qwen3.5 35B A3B";
+    readonly description: "Open-weight Qwen3.5 native vision-language MoE model (35B total, 3B active) with hybrid linear attention.";
+    readonly family: "alibaba";
+    readonly releasedAt: Date;
+    readonly providers: [{
+        readonly providerId: "alibaba";
+        readonly externalId: "qwen3.5-35b-a3b";
+        readonly inputPrice: "0.25e-6";
+        readonly outputPrice: "2e-6";
+        readonly regions: [{
+            readonly id: "singapore";
+        }];
+        readonly requestPrice: "0";
+        readonly contextSize: 262144;
+        readonly maxOutput: 65536;
+        readonly maxTemperature: 1;
+        readonly reasoning: true;
+        readonly reasoningEfforts: ["none", "minimal", "low", "medium", "high", "xhigh", "max"];
+        readonly reasoningMaxTokens: true;
+        readonly streaming: true;
+        readonly vision: true;
+        readonly tools: true;
+        readonly jsonOutput: true;
+        readonly supportedParameters: ["temperature", "max_tokens", "top_p", "frequency_penalty", "presence_penalty", "stop", "stream", "response_format", "tools", "reasoning_effort"];
+    }];
 }, {
     readonly id: "qwen3.6-35b-a3b";
     readonly name: "Qwen3.6 35B A3B";

package/dist/models/alibaba.js

@@ -2103,6 +2103,7 @@ export const alibabaModels = [
             {
                 providerId: "deepinfra",
                 externalId: "Qwen/Qwen3.8-27B",
+                deactivatedAt: new Date("2026-09-23"),
                 inputPrice: "0.2e-6",
                 cachedInputPrice: "0.05e-6",
                 outputPrice: "2.5e-6",
@@ -2158,6 +2159,7 @@ export const alibabaModels = [
                 maxOutput: 131072,
                 quantization: "fp4",
                 maxTemperature: 1,
+                minCacheableTokens: 4096,
                 streaming: true,
                 reasoning: true,
                 reasoningEfforts: ["minimal", "low", "medium", "high", "xhigh", "max"],
@@ -3782,6 +3784,175 @@ export const alibabaModels = [
             },
         ],
     },
+    {
+        id: "qwen3.5-122b-a10b",
+        name: "Qwen3.5 122B A10B",
+        description: "Open-weight Qwen3.5 native vision-language MoE model (122B total, 10B active) with hybrid linear attention.",
+        family: "alibaba",
+        releasedAt: new Date("2026-02-26"),
+        providers: [
+            {
+                providerId: "alibaba",
+                externalId: "qwen3.5-122b-a10b",
+                inputPrice: "0.4e-6",
+                outputPrice: "3.2e-6",
+                regions: [{ id: "singapore" }],
+                requestPrice: "0",
+                contextSize: 262144,
+                maxOutput: 65536,
+                maxTemperature: 1,
+                reasoning: true,
+                reasoningEfforts: [
+                    "none",
+                    "minimal",
+                    "low",
+                    "medium",
+                    "high",
+                    "xhigh",
+                    "max",
+                ],
+                reasoningMaxTokens: true,
+                streaming: true,
+                vision: true,
+                tools: true,
+                jsonOutput: true,
+                supportedParameters: [
+                    "temperature",
+                    "max_tokens",
+                    "top_p",
+                    "frequency_penalty",
+                    "presence_penalty",
+                    "stop",
+                    "stream",
+                    "response_format",
+                    "tools",
+                    "reasoning_effort",
+                ],
+            },
+            {
+                providerId: "novita",
+                externalId: "qwen/qwen3.5-122b-a10b",
+                inputPrice: "0.4e-6",
+                outputPrice: "3.2e-6",
+                requestPrice: "0",
+                contextSize: 262144,
+                maxOutput: 65536,
+                quantization: "bf16",
+                maxTemperature: 1,
+                reasoning: true,
+                streaming: true,
+                vision: true,
+                requiresBase64Images: true,
+                tools: true,
+                jsonOutput: true,
+                supportedParameters: [
+                    "temperature",
+                    "max_tokens",
+                    "top_p",
+                    "frequency_penalty",
+                    "presence_penalty",
+                    "stop",
+                    "stream",
+                    "response_format",
+                    "tools",
+                ],
+            },
+        ],
+    },
+    {
+        id: "qwen3.5-27b",
+        name: "Qwen3.5 27B",
+        description: "Open-weight dense Qwen3.5 native vision-language model with linear attention for fast responses.",
+        family: "alibaba",
+        releasedAt: new Date("2026-02-26"),
+        providers: [
+            {
+                providerId: "alibaba",
+                externalId: "qwen3.5-27b",
+                inputPrice: "0.3e-6",
+                outputPrice: "2.4e-6",
+                regions: [{ id: "singapore" }],
+                requestPrice: "0",
+                contextSize: 262144,
+                maxOutput: 65536,
+                maxTemperature: 1,
+                reasoning: true,
+                reasoningEfforts: [
+                    "none",
+                    "minimal",
+                    "low",
+                    "medium",
+                    "high",
+                    "xhigh",
+                    "max",
+                ],
+                reasoningMaxTokens: true,
+                streaming: true,
+                vision: true,
+                tools: true,
+                jsonOutput: true,
+                supportedParameters: [
+                    "temperature",
+                    "max_tokens",
+                    "top_p",
+                    "frequency_penalty",
+                    "presence_penalty",
+                    "stop",
+                    "stream",
+                    "response_format",
+                    "tools",
+                    "reasoning_effort",
+                ],
+            },
+        ],
+    },
+    {
+        id: "qwen3.5-35b-a3b",
+        name: "Qwen3.5 35B A3B",
+        description: "Open-weight Qwen3.5 native vision-language MoE model (35B total, 3B active) with hybrid linear attention.",
+        family: "alibaba",
+        releasedAt: new Date("2026-02-26"),
+        providers: [
+            {
+                providerId: "alibaba",
+                externalId: "qwen3.5-35b-a3b",
+                inputPrice: "0.25e-6",
+                outputPrice: "2e-6",
+                regions: [{ id: "singapore" }],
+                requestPrice: "0",
+                contextSize: 262144,
+                maxOutput: 65536,
+                maxTemperature: 1,
+                reasoning: true,
+                reasoningEfforts: [
+                    "none",
+                    "minimal",
+                    "low",
+                    "medium",
+                    "high",
+                    "xhigh",
+                    "max",
+                ],
+                reasoningMaxTokens: true,
+                streaming: true,
+                vision: true,
+                tools: true,
+                jsonOutput: true,
+                supportedParameters: [
+                    "temperature",
+                    "max_tokens",
+                    "top_p",
+                    "frequency_penalty",
+                    "presence_penalty",
+                    "stop",
+                    "stream",
+                    "response_format",
+                    "tools",
+                    "reasoning_effort",
+                ],
+            },
+        ],
+    },
     {
         id: "qwen3.6-35b-a3b",
         name: "Qwen3.6 35B A3B",