@juspay/neurolink 12.32.3 → 12.34.0

package/dist/constants/contextWindows.js

@@ -83,6 +83,10 @@ export const MODEL_CONTEXT_WINDOWS = {
     voyage: {
         // Voyage embeddings: max input tokens vary 16K-32K per model
         _default: 32_000,
+        "voyage-4": 32_000,
+        "voyage-4-large": 32_000,
+        "voyage-4-lite": 32_000,
+        "voyage-code-4": 32_000,
         "voyage-3.5": 32_000,
         "voyage-3.5-lite": 32_000,
         "voyage-3-large": 32_000,
@@ -111,6 +115,10 @@ export const MODEL_CONTEXT_WINDOWS = {
     },
     anthropic: {
         _default: 200_000,
+        // Claude 5.5 / 5.1 (Sep 2026) — 1M context window, confirmed on
+        // platform.claude.com/docs/en/about-claude/models/overview
+        "claude-opus-5-5": 1_000_000,
+        "claude-fable-5-1": 1_000_000,
         // Claude 5 (mid 2026) — 1M context window
         "claude-sonnet-5": 1_000_000,
         // Claude 4.6 (Feb 2026) — 1M context window
@@ -134,6 +142,11 @@ export const MODEL_CONTEXT_WINDOWS = {
     },
     openai: {
         _default: 128_000,
+        // GPT-6 family (Sep 2026) — 1.05M context, per
+        // developers.openai.com/api/docs/models
+        "gpt-6-astra": 1_050_000,
+        "gpt-6-sol": 1_050_000,
+        "gpt-6-luna": 1_050_000,
         // GPT-5.4 family — 1.05M context
         "gpt-5.4": 1_050_000,
         "gpt-5.4-mini": 400_000,
@@ -202,6 +215,8 @@ export const MODEL_CONTEXT_WINDOWS = {
     vertex: {
         _default: 1_048_576,
         // Claude on Vertex
+        "claude-opus-5-5": 1_000_000,
+        "claude-fable-5-1": 1_000_000,
         "claude-sonnet-5": 1_000_000,
         "claude-opus-4-6": 1_000_000,
         "claude-sonnet-4-6": 1_000_000,
@@ -238,6 +253,9 @@ export const MODEL_CONTEXT_WINDOWS = {
     },
     bedrock: {
         _default: 200_000,
+        // Claude 5.5 / 5.1 (Sep 2026)
+        "anthropic.claude-opus-5-5": 1_000_000,
+        "anthropic.claude-fable-5-1": 1_000_000,
         // Claude 4.6
         "anthropic.claude-opus-4-6-v1": 1_000_000,
         "anthropic.claude-sonnet-4-6": 1_000_000,
@@ -263,6 +281,16 @@ export const MODEL_CONTEXT_WINDOWS = {
     },
     azure: {
         _default: 128_000,
+        // GPT-6 family (Sep 2026) — 1.05M context, per
+        // learn.microsoft.com/azure/ai-foundry/openai/concepts/models
+        "gpt-6-astra": 1_050_000,
+        "gpt-6-sol": 1_050_000,
+        "gpt-6-luna": 1_050_000,
+        // GPT-5.6 / GPT-5.5 — Azure-only as of Sep 2026, same 1.05M context
+        "gpt-5.6-sol": 1_050_000,
+        "gpt-5.6-terra": 1_050_000,
+        "gpt-5.6-luna": 1_050_000,
+        "gpt-5.5": 1_050_000,
         // GPT-5.4
         "gpt-5.4": 1_050_000,
         "gpt-5.4-mini": 400_000,

package/dist/constants/enums.d.ts

@@ -180,6 +180,9 @@ export declare enum BedrockModels {
  * Supported Models for OpenAI
  */
 export declare enum OpenAIModels {
+    GPT_6_ASTRA = "gpt-6-astra",
+    GPT_6_SOL = "gpt-6-sol",
+    GPT_6_LUNA = "gpt-6-luna",
     GPT_5_3_CODEX = "gpt-5.3-codex",
     GPT_5_4 = "gpt-5.4",
     GPT_5_4_MINI = "gpt-5.4-mini",
@@ -228,6 +231,13 @@ export declare enum OpenAIModels {
  * Note: Azure uses deployment names, these are model identifiers
  */
 export declare enum AzureOpenAIModels {
+    GPT_6_ASTRA = "gpt-6-astra",
+    GPT_6_SOL = "gpt-6-sol",
+    GPT_6_LUNA = "gpt-6-luna",
+    GPT_5_6_SOL = "gpt-5.6-sol",
+    GPT_5_6_TERRA = "gpt-5.6-terra",
+    GPT_5_6_LUNA = "gpt-5.6-luna",
+    GPT_5_5 = "gpt-5.5",
     GPT_5_2 = "gpt-5.2",
     GPT_5_2_CHAT = "gpt-5.2-chat",
     GPT_5_2_PRO = "gpt-5.2-pro",
@@ -355,6 +365,8 @@ export declare enum AnthropicModels {
     CLAUDE_OPUS_5 = "claude-opus-5",
     CLAUDE_SONNET_5 = "claude-sonnet-5",
     CLAUDE_FABLE_5 = "claude-fable-5",
+    CLAUDE_OPUS_5_5 = "claude-opus-5-5",
+    CLAUDE_FABLE_5_1 = "claude-fable-5-1",
     CLAUDE_OPUS_4_8 = "claude-opus-4-8",
     CLAUDE_OPUS_4_7 = "claude-opus-4-7",
     CLAUDE_OPUS_4_6 = "claude-opus-4-6",
@@ -650,6 +662,14 @@ export declare enum CohereModels {
  * @see https://docs.voyageai.com/docs/embeddings
  */
 export declare enum VoyageModels {
+    /** Voyage 4 — latest general-purpose flagship */
+    VOYAGE_4 = "voyage-4",
+    /** Voyage 4 Large — largest / highest quality */
+    VOYAGE_4_LARGE = "voyage-4-large",
+    /** Voyage 4 Lite — smaller / cheaper */
+    VOYAGE_4_LITE = "voyage-4-lite",
+    /** Voyage Code 4 — code-tuned */
+    VOYAGE_CODE_4 = "voyage-code-4",
     /** Voyage 3.5 — latest general-purpose (default) */
     VOYAGE_3_5 = "voyage-3.5",
     /** Voyage 3.5 Lite — smaller / cheaper */
@@ -1042,6 +1062,7 @@ export declare enum UpstageModels {
     SYN_PRO_251021 = "syn-pro-251021"
 }
 export declare enum XaiModels {
+    GROK_4_7 = "grok-4.7",
     GROK_4_6 = "grok-4.6",
     GROK_4_5 = "grok-4.5",
     GROK_4_3 = "grok-4.3",

package/dist/constants/enums.js

@@ -257,6 +257,11 @@ export var BedrockModels;
  */
 export var OpenAIModels;
 (function (OpenAIModels) {
+    // GPT-6 Series (Released September 2026) - current flagship family, per
+    // developers.openai.com/api/docs/models
+    OpenAIModels["GPT_6_ASTRA"] = "gpt-6-astra";
+    OpenAIModels["GPT_6_SOL"] = "gpt-6-sol";
+    OpenAIModels["GPT_6_LUNA"] = "gpt-6-luna";
     // GPT-5.3 Series (Released February 2026) - Latest coding models
     OpenAIModels["GPT_5_3_CODEX"] = "gpt-5.3-codex";
     // GPT-5.4 Series (Released March 2026) - Latest flagship models
@@ -318,6 +323,18 @@ export var OpenAIModels;
  */
 export var AzureOpenAIModels;
 (function (AzureOpenAIModels) {
+    // GPT-6 Series (September 2026) - current flagship family, confirmed on
+    // Azure OpenAI (learn.microsoft.com/azure/ai-foundry/openai/concepts/models)
+    AzureOpenAIModels["GPT_6_ASTRA"] = "gpt-6-astra";
+    AzureOpenAIModels["GPT_6_SOL"] = "gpt-6-sol";
+    AzureOpenAIModels["GPT_6_LUNA"] = "gpt-6-luna";
+    // GPT-5.6 Series (Azure-only as of Sep 2026 — retired from the OpenAI-direct
+    // catalog, still documented on Azure OpenAI)
+    AzureOpenAIModels["GPT_5_6_SOL"] = "gpt-5.6-sol";
+    AzureOpenAIModels["GPT_5_6_TERRA"] = "gpt-5.6-terra";
+    AzureOpenAIModels["GPT_5_6_LUNA"] = "gpt-5.6-luna";
+    // GPT-5.5 (Azure-only as of Sep 2026 — same situation as GPT-5.6 above)
+    AzureOpenAIModels["GPT_5_5"] = "gpt-5.5";
     // GPT-5.2 Series (Latest - December 2025)
     AzureOpenAIModels["GPT_5_2"] = "gpt-5.2";
     AzureOpenAIModels["GPT_5_2_CHAT"] = "gpt-5.2-chat";
@@ -482,6 +499,10 @@ export var AnthropicModels;
     AnthropicModels["CLAUDE_OPUS_5"] = "claude-opus-5";
     AnthropicModels["CLAUDE_SONNET_5"] = "claude-sonnet-5";
     AnthropicModels["CLAUDE_FABLE_5"] = "claude-fable-5";
+    // Claude 5.5 / 5.1 Series (September 2026) — confirmed on
+    // platform.claude.com/docs/en/about-claude/models/overview
+    AnthropicModels["CLAUDE_OPUS_5_5"] = "claude-opus-5-5";
+    AnthropicModels["CLAUDE_FABLE_5_1"] = "claude-fable-5-1";
     // Claude 4.7 / 4.8 Series
     AnthropicModels["CLAUDE_OPUS_4_8"] = "claude-opus-4-8";
     AnthropicModels["CLAUDE_OPUS_4_7"] = "claude-opus-4-7";
@@ -868,6 +889,14 @@ export var CohereModels;
  */
 export var VoyageModels;
 (function (VoyageModels) {
+    /** Voyage 4 — latest general-purpose flagship */
+    VoyageModels["VOYAGE_4"] = "voyage-4";
+    /** Voyage 4 Large — largest / highest quality */
+    VoyageModels["VOYAGE_4_LARGE"] = "voyage-4-large";
+    /** Voyage 4 Lite — smaller / cheaper */
+    VoyageModels["VOYAGE_4_LITE"] = "voyage-4-lite";
+    /** Voyage Code 4 — code-tuned */
+    VoyageModels["VOYAGE_CODE_4"] = "voyage-code-4";
     /** Voyage 3.5 — latest general-purpose (default) */
     VoyageModels["VOYAGE_3_5"] = "voyage-3.5";
     /** Voyage 3.5 Lite — smaller / cheaper */
@@ -1290,6 +1319,7 @@ export var UpstageModels;
 })(UpstageModels || (UpstageModels = {}));
 export var XaiModels;
 (function (XaiModels) {
+    XaiModels["GROK_4_7"] = "grok-4.7";
     XaiModels["GROK_4_6"] = "grok-4.6";
     XaiModels["GROK_4_5"] = "grok-4.5";
     XaiModels["GROK_4_3"] = "grok-4.3";

package/dist/constants/tokens.d.ts

@@ -71,6 +71,9 @@ export declare const PROVIDER_TOKEN_LIMITS: {
     };
     /** OpenAI model limits */
     readonly OPENAI: {
+        readonly "gpt-6-astra": 128000;
+        readonly "gpt-6-sol": 128000;
+        readonly "gpt-6-luna": 128000;
         readonly "gpt-5.4": 128000;
         readonly "gpt-5.4-mini": 128000;
         readonly "gpt-5.4-nano": 128000;

package/dist/constants/tokens.js

@@ -74,6 +74,9 @@ export const PROVIDER_TOKEN_LIMITS = {
     },
     /** OpenAI model limits */
     OPENAI: {
+        "gpt-6-astra": 128_000,
+        "gpt-6-sol": 128_000,
+        "gpt-6-luna": 128_000,
         "gpt-5.4": 128_000,
         "gpt-5.4-mini": 128_000,
         "gpt-5.4-nano": 128_000,

package/dist/evaluation/EvaluatorFactory.js

@@ -57,7 +57,7 @@ export class EvaluatorFactory extends BaseFactory {
             const mergedConfig = {
                 threshold: 7,
                 evaluationStrategy: "ragas",
-                evaluationModel: process.env.NEUROLINK_RAGAS_EVALUATION_MODEL || "gemini-1.5-flash",
+                evaluationModel: process.env.NEUROLINK_RAGAS_EVALUATION_MODEL || "gemini-2.5-flash",
                 provider: process.env.NEUROLINK_RAGAS_EVALUATION_PROVIDER || "vertex",
                 ...config,
             };
@@ -95,7 +95,7 @@ export class EvaluatorFactory extends BaseFactory {
             const mergedConfig = {
                 threshold: 5,
                 evaluationStrategy: "ragas",
-                evaluationModel: process.env.NEUROLINK_RAGAS_EVALUATION_MODEL || "gemini-1.5-flash",
+                evaluationModel: process.env.NEUROLINK_RAGAS_EVALUATION_MODEL || "gemini-2.5-flash",
                 provider: process.env.NEUROLINK_RAGAS_EVALUATION_PROVIDER || "vertex",
                 ...config,
             };
@@ -114,7 +114,7 @@ export class EvaluatorFactory extends BaseFactory {
             const mergedConfig = {
                 threshold: 6,
                 evaluationStrategy: "ragas",
-                evaluationModel: "gemini-1.5-flash",
+                evaluationModel: "gemini-2.5-flash",
                 provider: "vertex",
                 ...config,
             };

package/dist/evaluation/EvaluatorRegistry.js

@@ -73,7 +73,7 @@ export class EvaluatorRegistry extends BaseRegistry {
             name: "RAGAS Evaluator",
             description: "RAGAS-style LLM-as-judge evaluation with relevance, accuracy, and completeness metrics",
             requiresLLM: true,
-            defaultModel: "gemini-1.5-flash",
+            defaultModel: "gemini-2.5-flash",
             defaultProvider: "vertex",
             version: "1.0.0",
             features: [

package/dist/evaluation/ragasEvaluator.js

@@ -18,7 +18,7 @@ export class RAGASEvaluator {
         this.evaluationModel =
             evaluationModel ||
                 process.env.NEUROLINK_RAGAS_EVALUATION_MODEL ||
-                "gemini-1.5-flash";
+                "gemini-2.5-flash";
         this.providerName =
             providerName ||
                 process.env.NEUROLINK_RAGAS_EVALUATION_PROVIDER ||

package/dist/middleware/utils/guardrailsUtils.js

@@ -113,7 +113,7 @@ params, config) {
  */
 export async function performPrecallEvaluation(config, userInput) {
     try {
-        const provider = await AIProviderFactory.createProvider(config.provider || "google-ai", config.evaluationModel || "gemini-1.5-flash");
+        const provider = await AIProviderFactory.createProvider(config.provider || "google-ai", config.evaluationModel || "gemini-2.5-flash");
         const evaluationPrompt = config.evaluationPrompt || DEFAULT_EVALUATION_PROMPT;
         const prompt = evaluationPrompt.replace("{USER_INPUT}", userInput);
         const result = await provider.generate({

package/dist/models/manifests/anthropic.js

@@ -23,11 +23,43 @@ export const anthropicManifest = {
         },
     ],
     models: {
+        "claude-opus-5-5": {
+            aliases: ["opus-5.5"],
+            displayName: "Claude Opus 5.5",
+            contextWindow: 1_000_000,
+            maxOutputTokens: 128_000,
+            // No pricingPerMTok DELIBERATELY: PRICING.anthropic already carries
+            // real rates for this id (input 4.0/output 20.0 per MTok) — see the
+            // claude-sonnet-5 comment below for why this defers to that table.
+            vision: true,
+            functionCalling: true,
+            reasoning: true,
+            jsonMode: true,
+            samplingParams: false, // matches SAMPLING_PARAM_REJECTING_FAMILIES /opus[-_.]?5(?![0-9])/i
+        },
+        "claude-fable-5-1": {
+            aliases: ["fable-5.1"],
+            displayName: "Claude Fable 5.1",
+            contextWindow: 1_000_000,
+            maxOutputTokens: 128_000,
+            // No pricingPerMTok DELIBERATELY: PRICING.anthropic already carries
+            // real rates for this id (input 10.0/output 50.0 per MTok) — see the
+            // claude-sonnet-5 comment below for why this defers to that table.
+            vision: true,
+            functionCalling: true,
+            reasoning: true,
+            jsonMode: true,
+            samplingParams: false, // matches SAMPLING_PARAM_REJECTING_FAMILIES /fable/i
+        },
         "claude-sonnet-5": {
             aliases: ["sonnet-5", "claude-sonnet"],
             displayName: "Claude Sonnet 5",
             contextWindow: 1_000_000,
-            maxOutputTokens: 64_000,
+            // Was 64_000; corrected to 128_000 to match the official spec
+            // (platform.claude.com/docs/en/about-claude/models/overview) and the
+            // matching fix in getClaudeMaxOutputTokens (tokenLimits.ts), which this
+            // manifest's own header comment names as the authoritative source.
+            maxOutputTokens: 128_000,
             // No pricingPerMTok DELIBERATELY: PRICING.anthropic carries real
             // rates for this id, and findRates() is manifest-first with a legacy
             // fallback — omitting the rate here defers to that table instead of

package/dist/providers/catalog/api-route.json

@@ -85,7 +85,7 @@
         "tools": true,
         "toolsWithStreaming": true,
         "structuredOutput": true,
-        "structuredOutputWithTools": true,
+        "structuredOutputWithTools": false,
         "embeddings": false,
         "thinking": false
     },
@@ -123,7 +123,7 @@
         },
         "liveMatrix": {
             "date": "2026-09-17",
-            "result": "Verified live against https://global.api-route.com/v1: GET /v1/models returned 70 models. Chat completions, streaming, function calling, streamed function calling, and structured JSON output (json_object and json_schema) verified. Multimodal vision verified on claude-sonnet-4-6, claude-haiku-4-5, and gemini-3.8-flash. Invalid API key returns 401 (无效的令牌) and invalid model returns 503 (model_not_found)."
+            "result": "Verified live against https://global.api-route.com/v1: GET /v1/models returned 70 models. Chat completions, streaming, function calling, streamed function calling, and structured JSON output (json_object and json_schema) verified. Multimodal vision verified on claude-sonnet-4-6, claude-haiku-4-5, and gemini-3.8-flash. Invalid API key returns 401 (无效的令牌) and invalid model returns 503 (model_not_found). Tools and response_format were only probed SEPARATELY, never together in one request, so structuredOutputWithTools is corrected to false here pending a dedicated tools_plus_schema probe — the individual passes do not demonstrate the combined capability."
         },
         "addedInPR": "pending"
     }

package/dist/providers/catalog/loader.js

@@ -131,6 +131,11 @@ export function buildCatalogEntries() {
             ...(typeof entry.capabilities.tools === "boolean"
                 ? { supportsTools: entry.capabilities.tools }
                 : {}),
+            // Unlike `tools`, `capabilities.structuredOutputWithTools` is a plain
+            // required boolean in the schema (no "model-dependent" member), so this
+            // is a direct passthrough — `false` and `true` both carry meaning to
+            // ConfiguredOpenAICompatProvider.suppressResponseFormatWithTools().
+            supportsStructuredOutputWithTools: entry.capabilities.structuredOutputWithTools,
         };
         const { baseURLTemplate } = entry.wire;
         if (baseURLTemplate) {

package/dist/providers/catalog/mistral.json

@@ -208,7 +208,7 @@
         "tools": true,
         "toolsWithStreaming": true,
         "structuredOutput": true,
-        "structuredOutputWithTools": true,
+        "structuredOutputWithTools": false,
         "embeddings": false,
         "thinking": false
     },
@@ -245,7 +245,7 @@
         },
         "liveMatrix": {
             "date": "2026-08-30",
-            "result": "17 ids answer normally. A chat probe alone was not enough to judge the rest, because Mistral rejects every non-chat model with the same generic 'Invalid model' text: cross-referencing the roster shows mistral-embed, codestral-embed, mistral-ocr-latest, mistral-ocr-2512, voxtral-mini-latest and voxtral-mini-2602 are all still listed and alive — they are embedding, OCR and audio models that a chat endpoint is right to refuse — so they keep production status. An id is marked retired only when BOTH signals agree: absent from the roster AND rejected by a live chat call. pixtral-12b is the one id where they disagree — the roster omits it but a real call returns 200, so it was moved back to production, on the rule that a live 200 outranks a roster omission. mistral-large-latest and mistral-large-2512 answer 403 'not available in your subscription tier': a plan limit on this account, not a retirement, so they stay production. visionModel is pinned to mistral-medium-latest because the otherwise-first vision id, mistral-large-latest, is behind that plan gate."
+            "result": "17 ids answer normally. A chat probe alone was not enough to judge the rest, because Mistral rejects every non-chat model with the same generic 'Invalid model' text: cross-referencing the roster shows mistral-embed, codestral-embed, mistral-ocr-latest, mistral-ocr-2512, voxtral-mini-latest and voxtral-mini-2602 are all still listed and alive — they are embedding, OCR and audio models that a chat endpoint is right to refuse — so they keep production status. An id is marked retired only when BOTH signals agree: absent from the roster AND rejected by a live chat call. pixtral-12b is the one id where they disagree — the roster omits it but a real call returns 200, so it was moved back to production, on the rule that a live 200 outranks a roster omission. mistral-large-latest and mistral-large-2512 answer 403 'not available in your subscription tier': a plan limit on this account, not a retirement, so they stay production. visionModel is pinned to mistral-medium-latest because the otherwise-first vision id, mistral-large-latest, is behind that plan gate. Combined tools plus response_format probes returned 429 twice, which is not successful capability evidence, so structuredOutputWithTools remains false pending a successful combined-request probe."
         },
         "addedInPR": "https://github.com/juspay/neurolink/pull/1587"
     }

package/dist/providers/catalog/xai.json

@@ -9,10 +9,24 @@
     },
     "models": {
         "default": "grok-4.6",
-        "fallbacks": ["grok-4.6", "grok-4.5", "grok-4.3", "grok-3", "grok-3-mini"],
+        "fallbacks": [
+            "grok-4.6",
+            "grok-4.5",
+            "grok-4.3",
+            "grok-3",
+            "grok-3-mini",
+            "grok-4.7"
+        ],
         "defaultContextWindow": 131072,
         "defaultMaxOutputTokens": 4096,
         "catalog": {
+            "grok-4.7": {
+                "enumMember": "GROK_4_7",
+                "contextWindow": 500000,
+                "vision": true,
+                "status": "production",
+                "description": "Grok 4.7 — xAI's newest flagship as of 2026-09-27 (text + image input); not yet the SDK default, see docs.x.ai/developers/grok-4-7"
+            },
             "grok-4.6": {
                 "enumMember": "GROK_4_6",
                 "contextWindow": 131072,
@@ -71,7 +85,7 @@
             }
         },
         "visionModel": "grok-4.6",
-        "topModels": ["grok-4.6", "grok-4.5", "grok-4.3", "grok-3"]
+        "topModels": ["grok-4.6", "grok-4.5", "grok-4.3", "grok-3", "grok-4.7"]
     },
     "capabilities": {
         "text": true,
@@ -114,7 +128,7 @@
         },
         "liveMatrix": {
             "date": "2026-08-30",
-            "result": "grok-2-latest, grok-2-vision-latest and grok-beta are rejected as invalid models. The grok-4.x line replaces them: 4.3/4.5/4.6 all answer, and all three read a 64x64 PNG correctly, so vision moved there from the retired grok-2-vision-latest. An earlier probe with an 8x8 PNG reported no vision — that was xAI answering 'Invalid PNG image', a complaint about the file rather than the capability. grok-3 and grok-3-mini are kept because they still serve, even though xAI's /models roster omits them: roster absence is not evidence of death here."
+            "result": "grok-2-latest, grok-2-vision-latest and grok-beta are rejected as invalid models. The grok-4.x line replaces them: 4.3/4.5/4.6 all answer, and all three read a 64x64 PNG correctly, so vision moved there from the retired grok-2-vision-latest. An earlier probe with an 8x8 PNG reported no vision — that was xAI answering 'Invalid PNG image', a complaint about the file rather than the capability. grok-3 and grok-3-mini are kept because they still serve, even though xAI's /models roster omits them: roster absence is not evidence of death here. grok-4.7 added 2026-09-27 from docs.x.ai/developers/grok-4-7 and docs.x.ai/developers/pricing (official docs review only — no xAI API key was available in this environment, so unlike the rest of this file it has not been roster/chat-probed live; context 500,000 tokens and text+image input are as published, and it is deliberately left with no pricingPerMTok, matching grok-4.6/4.5/4.3, because its real rate is tiered by prompt length ($2/$0.50/$6 per 1M below 200k tokens, $4/$1/$12 at or above) and the schema's pricingPerMTok field can only hold one flat rate)."
         },
         "addedInPR": "https://github.com/juspay/neurolink/pull/1587"
     }

package/dist/providers/configuredOpenAICompat.d.ts

@@ -9,8 +9,10 @@ import { OpenAIChatCompletionsProvider } from "./openaiChatCompletionsBase.js";
  * OPENAI_COMPAT_CATALOG in openaiCompatCatalog.ts for the entries.
  *
  * If a provider needs a real hook override (adjustRequestBody,
- * adjustBodyAfter400, getChatCompletionsURL, getAuthHeaders,
- * suppressResponseFormatWithTools, ...) it does NOT belong in the catalog —
+ * adjustBodyAfter400, getChatCompletionsURL, getAuthHeaders, ...) that isn't
+ * expressible as one of the catalog's data-driven quirks (messageContentFormat,
+ * responseFormatDowngrade, replayReasoningContent,
+ * supportsStructuredOutputWithTools, ...) it does NOT belong in the catalog —
  * write a dedicated subclass instead (see azureOpenai.ts).
  *
  * The exception is a WIRE DIALECT: a vendor that speaks OpenAI for ordinary
@@ -55,6 +57,22 @@ export declare class ConfiguredOpenAICompatProvider extends OpenAIChatCompletion
      * construction: BaseProvider merely closes over it.
      */
     supportsTools(): boolean;
+    /**
+     * The base class drops `response_format` from any request carrying tools,
+     * because a generic OpenAI-compatible backend may honour it over tool
+     * calling and answer with final-shape JSON on step one instead of running
+     * the agentic loop. For a catalog entry that is not a guess — the flag
+     * comes from a live wire probe (`capabilities.structuredOutputWithTools`)
+     * — so an entry that declares the combination supported sends both and
+     * gets the object from the vendor's own constrained decoding in ONE
+     * request, instead of falling back to a tool-free re-ask.
+     *
+     * Conservative by default: only an explicit `true` opts in. A vendor that
+     * declares support but rejects the combination at runtime is still caught
+     * by `isToolsSchemaConflictError` and retried without structured output,
+     * so a stale flag costs one request rather than the turn.
+     */
+    protected suppressResponseFormatWithTools(): boolean;
     protected adjustRequestBody(body: OpenAICompatChatRequest, modelId: string): OpenAICompatChatRequest;
     protected adjustResponseFormat(rf: OpenAICompatResponseFormat | undefined, modelId: string): OpenAICompatResponseFormat | undefined;
     protected replayReasoningContent(): boolean;

package/dist/providers/configuredOpenAICompat.js

@@ -31,8 +31,10 @@ const flattenMessageContent = (content) => {
  * OPENAI_COMPAT_CATALOG in openaiCompatCatalog.ts for the entries.
  *
  * If a provider needs a real hook override (adjustRequestBody,
- * adjustBodyAfter400, getChatCompletionsURL, getAuthHeaders,
- * suppressResponseFormatWithTools, ...) it does NOT belong in the catalog —
+ * adjustBodyAfter400, getChatCompletionsURL, getAuthHeaders, ...) that isn't
+ * expressible as one of the catalog's data-driven quirks (messageContentFormat,
+ * responseFormatDowngrade, replayReasoningContent,
+ * supportsStructuredOutputWithTools, ...) it does NOT belong in the catalog —
  * write a dedicated subclass instead (see azureOpenai.ts).
  *
  * The exception is a WIRE DIALECT: a vendor that speaks OpenAI for ordinary
@@ -110,6 +112,24 @@ export class ConfiguredOpenAICompatProvider extends OpenAIChatCompletionsProvide
         }
         return super.supportsTools();
     }
+    /**
+     * The base class drops `response_format` from any request carrying tools,
+     * because a generic OpenAI-compatible backend may honour it over tool
+     * calling and answer with final-shape JSON on step one instead of running
+     * the agentic loop. For a catalog entry that is not a guess — the flag
+     * comes from a live wire probe (`capabilities.structuredOutputWithTools`)
+     * — so an entry that declares the combination supported sends both and
+     * gets the object from the vendor's own constrained decoding in ONE
+     * request, instead of falling back to a tool-free re-ask.
+     *
+     * Conservative by default: only an explicit `true` opts in. A vendor that
+     * declares support but rejects the combination at runtime is still caught
+     * by `isToolsSchemaConflictError` and retried without structured output,
+     * so a stale flag costs one request rather than the turn.
+     */
+    suppressResponseFormatWithTools() {
+        return this.entry.supportsStructuredOutputWithTools !== true;
+    }
     adjustRequestBody(body, modelId) {
         const adjusted = super.adjustRequestBody(body, modelId);
         if (this.entry.messageContentFormat !== "string") {

package/dist/providers/googleVertex/client.js

@@ -6480,8 +6480,6 @@ export class GoogleVertexProvider extends BaseProvider {
                 "gemini-2.5-flash-lite",
                 "gemini-2.0-flash-001",
                 "gemini-2.0-flash-lite",
-                "gemini-1.5-pro",
-                "gemini-1.5-flash",
             ],
             claude: [
                 "claude-sonnet-4-5@20250929",

package/dist/providers/litellm/client.js

@@ -13,7 +13,7 @@ import { createTimeoutController } from "../../utils/timeout.js";
 import { stripTrailingSlash } from "../openaiChatCompletionsClient.js";
 import { OpenAIChatCompletionsProvider } from "../openaiChatCompletionsBase.js";
 const streamTracer = trace.getTracer("neurolink.provider.litellm");
-const FALLBACK_LITELLM_MODEL = "openai/gpt-4o-mini";
+const FALLBACK_LITELLM_MODEL = "openai/gpt-5.4-mini";
 const getLiteLLMConfig = () => ({
     baseURL: process.env.LITELLM_BASE_URL || "http://localhost:4000",
     apiKey: process.env.LITELLM_API_KEY || "sk-anything",
@@ -243,7 +243,7 @@ export class LiteLLMProvider extends OpenAIChatCompletionsProvider {
             .map((m) => m.trim())
             .filter((m) => m.length > 0) || [
             "openai/gpt-4o",
-            "anthropic/claude-3-haiku",
+            "anthropic/claude-haiku-4-5-20251001",
             "meta-llama/llama-3.1-8b-instruct",
             "google/gemini-2.5-flash",
         ]);

package/dist/providers/openAI/client.js

@@ -44,7 +44,7 @@ const resolveOpenAIBaseURL = (credentialBaseURL, envBaseURL) => {
     return resolved;
 };
 const getOpenAIApiKey = () => validateApiKey(createOpenAIConfig());
-const getOpenAIModel = () => getProviderModel("OPENAI_MODEL", "gpt-4o");
+const getOpenAIModel = () => getProviderModel("OPENAI_MODEL", "gpt-5.4");
 const streamTracer = trace.getTracer("neurolink.provider.openai");
 /**
  * OpenAI Provider — direct HTTP, no AI SDK.

package/dist/providers/openRouter/client.js

@@ -278,7 +278,7 @@ export class OpenRouterProvider extends OpenAIChatCompletionsProvider {
             "openai/gpt-4-turbo",
             // Google models
             "google/gemini-2.0-flash",
-            "google/gemini-1.5-pro",
+            "google/gemini-2.5-pro",
             // Meta Llama models
             "meta-llama/llama-3.1-70b-instruct",
             "meta-llama/llama-3.1-8b-instruct",

package/dist/types/providerCatalog.d.ts

@@ -122,6 +122,16 @@ export type CatalogCapabilities = {
     tools: boolean | "model-dependent";
     toolsWithStreaming: boolean;
     structuredOutput: boolean;
+    /**
+     * Whether a live wire probe proved the vendor accepts native tool
+     * definitions and `response_format` in the SAME request (a 200, not a
+     * schema/tools conflict error). Maps to
+     * OpenAICompatCatalogEntry.supportsStructuredOutputWithTools; `true` lets
+     * ConfiguredOpenAICompatProvider send both instead of suppressing
+     * `response_format` on any request carrying tools (the conservative
+     * default for an unknown endpoint). Only an explicit `true` opts in —
+     * `false` is the same conservative behavior as leaving it unset.
+     */
     structuredOutputWithTools: boolean;
     embeddings: boolean;
     thinking: boolean;

package/dist/types/providers.d.ts

@@ -749,6 +749,26 @@ export type OpenAICompatCatalogEntry = {
      * same default every hand-written provider uses.
      */
     supportsTools?: boolean;
+    /**
+     * Whether the vendor accepts native tool definitions and `response_format`
+     * in the SAME request, from the catalog's
+     * `capabilities.structuredOutputWithTools`.
+     *
+     * The base class (`suppressResponseFormatWithTools`) suppresses
+     * `response_format` whenever tools are attached, because a generic
+     * OpenAI-compatible backend may honour it over tool calling and answer
+     * with final-shape JSON on step 1 instead of running the agentic loop.
+     * That is the right default for an unknown endpoint, but for a catalog
+     * entry it is not a guess: the flag is set from a live wire probe. `true`
+     * lets the request carry both, so the object comes back from the vendor's
+     * own constrained decoding in ONE request instead of a tool-free re-ask.
+     *
+     * Omitted or false keeps the conservative default. A vendor that turns
+     * out to reject the combination anyway is still caught by
+     * `isToolsSchemaConflictError` and retried without structured output, so
+     * a stale `true` costs one request, not the turn.
+     */
+    supportsStructuredOutputWithTools?: boolean;
     /**
      * The literal passed as ProviderFactory.registerProvider()'s defaultModel
      * argument (resolved before the provider is constructed). Preserves each

package/dist/utils/modelChoices.js

@@ -103,14 +103,6 @@ const TOP_MODELS_CONFIG = {
             model: GoogleAIModels.GEMINI_3_PRO_PREVIEW,
             description: "Latest preview",
         },
-        {
-            model: GoogleAIModels.GEMINI_1_5_PRO,
-            description: "Previous generation",
-        },
-        {
-            model: GoogleAIModels.GEMINI_1_5_FLASH,
-            description: "Legacy fast model",
-        },
     ],
     [AIProviderName.VERTEX]: [
         {
@@ -126,7 +118,6 @@ const TOP_MODELS_CONFIG = {
             model: VertexModels.GEMINI_2_0_FLASH,
             description: "Stable production model",
         },
-        { model: VertexModels.GEMINI_1_5_PRO, description: "Previous generation" },
         {
             model: VertexModels.CLAUDE_3_5_SONNET,
             description: "Claude 3.5 on Vertex",

package/dist/utils/pricing.js

@@ -685,6 +685,11 @@ const PRICING = {
         // Voyage bills per million input tokens — output dimension is the
         // embedding vector, not generated tokens. We charge to input only.
         _default: { input: 0.18 / 1_000_000, output: 0 },
+        // Voyage 4 family — docs.voyageai.com/docs/embeddings pricing table
+        "voyage-4": { input: 0.06 / 1_000_000, output: 0 },
+        "voyage-4-large": { input: 0.12 / 1_000_000, output: 0 },
+        "voyage-4-lite": { input: 0.02 / 1_000_000, output: 0 },
+        "voyage-code-4": { input: 0.12 / 1_000_000, output: 0 },
         "voyage-3.5": { input: 0.06 / 1_000_000, output: 0 },
         "voyage-3.5-lite": { input: 0.02 / 1_000_000, output: 0 },
         "voyage-3-large": { input: 0.18 / 1_000_000, output: 0 },