@juspay/neurolink 12.2.0 → 12.2.1

package/dist/constants/enums.d.ts

@@ -756,19 +756,15 @@ export declare enum GroqModels {
  * Cerebras inference models (wafer-scale, OpenAI-compatible API).
  * @see https://inference-docs.cerebras.ai/introduction
  *
- * Note the vendor's inconsistent id scheme: `llama3.1-8b` has no dash
- * after "llama", while `llama-3.3-70b` does — both are the DOCUMENTED
- * ids, not typos.
+ * Roster verified against a live authenticated `/v1/models` on 2026-08-27:
+ * only these two ids are served. The llama/qwen models the vendor docs
+ * once listed are retired and now 404.
  */
 export declare enum CerebrasModels {
-    /** Llama 3.3 70B — production default */
-    LLAMA_3_3_70B = "llama-3.3-70b",
-    /** Llama 3.1 8B — low-latency tier (vendor id has no dash after "llama") */
-    LLAMA_3_1_8B = "llama3.1-8b",
-    /** Qwen 3 32B */
-    QWEN_3_32B = "qwen-3-32b",
-    /** OpenAI GPT-OSS 120B (open-weight) */
-    GPT_OSS_120B = "gpt-oss-120b"
+    /** OpenAI GPT-OSS 120B (open-weight) — production default */
+    GPT_OSS_120B = "gpt-oss-120b",
+    /** Google Gemma 4 31B */
+    GEMMA_4_31B = "gemma-4-31b"
 }
 /**
  * Cohere Command + Embed models.

package/dist/constants/enums.js

@@ -1009,20 +1009,16 @@ export var GroqModels;
  * Cerebras inference models (wafer-scale, OpenAI-compatible API).
  * @see https://inference-docs.cerebras.ai/introduction
  *
- * Note the vendor's inconsistent id scheme: `llama3.1-8b` has no dash
- * after "llama", while `llama-3.3-70b` does — both are the DOCUMENTED
- * ids, not typos.
+ * Roster verified against a live authenticated `/v1/models` on 2026-08-27:
+ * only these two ids are served. The llama/qwen models the vendor docs
+ * once listed are retired and now 404.
  */
 export var CerebrasModels;
 (function (CerebrasModels) {
-    /** Llama 3.3 70B — production default */
-    CerebrasModels["LLAMA_3_3_70B"] = "llama-3.3-70b";
-    /** Llama 3.1 8B — low-latency tier (vendor id has no dash after "llama") */
-    CerebrasModels["LLAMA_3_1_8B"] = "llama3.1-8b";
-    /** Qwen 3 32B */
-    CerebrasModels["QWEN_3_32B"] = "qwen-3-32b";
-    /** OpenAI GPT-OSS 120B (open-weight) */
+    /** OpenAI GPT-OSS 120B (open-weight) — production default */
     CerebrasModels["GPT_OSS_120B"] = "gpt-oss-120b";
+    /** Google Gemma 4 31B */
+    CerebrasModels["GEMMA_4_31B"] = "gemma-4-31b";
 })(CerebrasModels || (CerebrasModels = {}));
 /**
  * Cohere Command + Embed models.

package/dist/factories/providerDescriptors.js

@@ -377,7 +377,7 @@ export const PROVIDER_DESCRIPTORS = [
             baseURL: "CEREBRAS_BASE_URL",
             model: "CEREBRAS_MODEL",
         },
-        defaultModel: CerebrasModels.LLAMA_3_3_70B,
+        defaultModel: CerebrasModels.GPT_OSS_120B,
         toolSupport: "native",
         localRuntime: false,
         healthCheck: "env-only",

package/dist/providers/openaiChatCompletionsClient.js

@@ -490,7 +490,12 @@ export const buildBody = (args) => {
     if (tools) {
         body.tools = tools;
     }
-    if (toolChoice !== undefined) {
+    // tool_choice is only meaningful alongside a non-empty tools array, and
+    // strict OpenAI-compatible backends (probed live on Cerebras 2026-08-27)
+    // reject it outright with 400 wrong_api_format when tools are absent:
+    // "'tool_choice' is only allowed when 'tools' are specified". Omitting it
+    // is behavior-identical on lenient backends.
+    if (toolChoice !== undefined && tools && tools.length > 0) {
         body.tool_choice = toolChoice;
     }
     if (responseFormat) {

package/dist/providers/openaiCompatCatalog.js

@@ -37,16 +37,11 @@ export const OPENAI_COMPAT_CATALOG = [
         defaultBaseURL: "https://api.cerebras.ai/v1",
         configOptions: createCerebrasConfig(),
         modelEnvVar: "CEREBRAS_MODEL",
-        defaultModel: CerebrasModels.LLAMA_3_3_70B,
-        registryDefaultModel: CerebrasModels.LLAMA_3_3_70B,
+        defaultModel: CerebrasModels.GPT_OSS_120B,
+        registryDefaultModel: CerebrasModels.GPT_OSS_120B,
         registryDefaultModelChecksEnvVar: true,
-        fallbackModelName: CerebrasModels.LLAMA_3_1_8B,
-        fallbackModels: [
-            CerebrasModels.LLAMA_3_3_70B,
-            CerebrasModels.LLAMA_3_1_8B,
-            CerebrasModels.QWEN_3_32B,
-            CerebrasModels.GPT_OSS_120B,
-        ],
+        fallbackModelName: CerebrasModels.GEMMA_4_31B,
+        fallbackModels: [CerebrasModels.GPT_OSS_120B, CerebrasModels.GEMMA_4_31B],
         errorRules: [
             {
                 // Probed live 2026-08-26: a bad key gets HTTP 401 with body

package/dist/utils/modelChoices.js

@@ -304,17 +304,12 @@ const TOP_MODELS_CONFIG = {
     ],
     [AIProviderName.CEREBRAS]: [
         {
-            model: CerebrasModels.LLAMA_3_3_70B,
-            description: "Recommended - Production default; wafer-scale speed",
-        },
-        {
-            model: CerebrasModels.LLAMA_3_1_8B,
-            description: "Lowest latency tier",
+            model: CerebrasModels.GPT_OSS_120B,
+            description: "Recommended - OpenAI GPT-OSS 120B (open-weight); wafer-scale speed",
         },
-        { model: CerebrasModels.QWEN_3_32B, description: "Qwen 3 32B" },
         {
-            model: CerebrasModels.GPT_OSS_120B,
-            description: "OpenAI GPT-OSS 120B (open-weight)",
+            model: CerebrasModels.GEMMA_4_31B,
+            description: "Google Gemma 4 31B",
         },
     ],
     [AIProviderName.COHERE]: [

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@juspay/neurolink",
-  "version": "12.2.0",
+  "version": "12.2.1",
   "packageManager": "pnpm@10.15.1",
   "description": "TypeScript AI SDK with 24+ LLM providers behind one consistent API. MCP-native (connect any MCP server), voice TTS/STT/realtime, RAG, agents, memory, context compaction. OpenAI · Anthropic · Gemini · Bedrock · Azure · Ollama · DeepSeek · NVIDIA NIM and more.",
   "author": {