@juspay/neurolink 12.38.0 → 12.39.0

package/dist/providers/catalog/chutes.json

@@ -0,0 +1,236 @@
+{
+    "$schema": "./provider-catalog.schema.json",
+    "id": "chutes",
+    "displayName": "Chutes",
+    "aliases": [],
+    "tier": 2,
+    "wire": {
+        "baseURL": "https://llm.chutes.ai/v1"
+    },
+    "models": {
+        "default": "moonshotai/Kimi-K2.6-TEE",
+        "fallbacks": [
+            "moonshotai/Kimi-K3-TEE",
+            "deepseek-ai/DeepSeek-V4-Flash-0731-TEE",
+            "zai-org/GLM-5.2-TEE",
+            "Qwen/Qwen3.5-397B-A17B-TEE",
+            "Qwen/Qwen3.6-27B-TEE",
+            "Qwen/Qwen3.8-27B-TEE",
+            "Qwen/Qwen3-235B-A22B-Thinking-2507-TEE",
+            "deepseek-ai/DeepSeek-V3.2-TEE",
+            "zai-org/GLM-5.1-TEE",
+            "google/gemma-4-31B-turbo-TEE",
+            "Qwen/Qwen3-32B-TEE",
+            "unsloth/Mistral-Nemo-Instruct-2407-TEE",
+            "Nemotron-3-Nano-Omni-30B-TEE"
+        ],
+        "defaultContextWindow": 262144,
+        "defaultMaxOutputTokens": 65535,
+        "catalog": {
+            "moonshotai/Kimi-K2.6-TEE": {
+                "contextWindow": 262144,
+                "maxOutputTokens": 65535,
+                "pricingPerMTok": {
+                    "input": 0.5,
+                    "output": 2.85,
+                    "cachedInput": 0.05
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Default — this is the model Chutes' own homepage code sample uses for the documented POST https://llm.chutes.ai/v1/chat/completions example (model, messages, stream:true, Authorization: Bearer $CHUTES_API_KEY), the strongest wire-shape evidence available without a key. credential-free roster probe: unauthenticated GET https://llm.chutes.ai/v1/models on 2026-09-28 (HTTP 200, 14 models). The public roster entry (root moonshotai/Kimi-K3, confidential_compute:true, owned_by vllm) lists context_length 262144, max_output_length 65535, input_modalities [text, image, video] and supported_features [json_mode, structured_outputs, tools, reasoning] — matched by the function-calling guide (https://chutes.ai/docs/guides/agents-and-tools, OpenAI-shaped tools/tool_choice/tool_call_id) and by the JSON-mode example on the same page (response_format:{type:\"json_object\"}). None of this was exercised end to end — no API key was used, so vision/tool/JSON-mode support and pricing are roster- and docs-verified only, not live-probed. Pricing ($0.50/$2.85 per MTok in/out, $0.05 cached) is read directly from the same GET /v1/models response and is consistent with https://chutes.ai/pricing's stated USD-per-1M-token PAYG billing."
+            },
+            "moonshotai/Kimi-K3-TEE": {
+                "contextWindow": 1048576,
+                "maxOutputTokens": 65535,
+                "pricingPerMTok": {
+                    "input": 3.0,
+                    "output": 15.0,
+                    "cachedInput": 0.3
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Moonshot Kimi K3 — largest context window in the roster (1,048,576 tokens) and the same declared feature set as the default (json_mode, structured_outputs, tools, reasoning) plus image/video input modalities per the public /v1/models entry. credential-free roster probe: unauthenticated GET https://llm.chutes.ai/v1/models on 2026-09-28 (HTTP 200, 14 models). Roster-listed only; not itself capability-probed (no key used); fallback."
+            },
+            "deepseek-ai/DeepSeek-V4-Flash-0731-TEE": {
+                "contextWindow": 1048576,
+                "maxOutputTokens": 131072,
+                "pricingPerMTok": {
+                    "input": 0.44,
+                    "output": 1.32,
+                    "cachedInput": 0.044
+                },
+                "vision": false,
+                "status": "production",
+                "description": "DeepSeek V4 Flash (0731 snapshot) — 1,048,576-token context window and the largest documented max_output_length in the roster (131,072). Text-only input per input_modalities. credential-free roster probe: unauthenticated GET https://llm.chutes.ai/v1/models on 2026-09-28 (HTTP 200, 14 models). Roster-listed only; not capability-probed; fallback."
+            },
+            "zai-org/GLM-5.2-TEE": {
+                "contextWindow": 1048576,
+                "maxOutputTokens": 65535,
+                "pricingPerMTok": {
+                    "input": 1.25,
+                    "output": 3.95,
+                    "cachedInput": 0.125
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Zhipu GLM-5.2 — 1,048,576-token context window, text-only input per input_modalities. credential-free roster probe: unauthenticated GET https://llm.chutes.ai/v1/models on 2026-09-28 (HTTP 200, 14 models). Roster-listed only; not capability-probed; fallback."
+            },
+            "Qwen/Qwen3.5-397B-A17B-TEE": {
+                "contextWindow": 262144,
+                "maxOutputTokens": 65536,
+                "pricingPerMTok": {
+                    "input": 0.45,
+                    "output": 3.0,
+                    "cachedInput": 0.045
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Qwen 3.5 397B-A17B (MoE) — image input documented in input_modalities alongside the full json_mode/tools/structured_outputs/reasoning feature set. credential-free roster probe: unauthenticated GET https://llm.chutes.ai/v1/models on 2026-09-28 (HTTP 200, 14 models). Roster-listed only; not capability-probed; fallback."
+            },
+            "Qwen/Qwen3.6-27B-TEE": {
+                "contextWindow": 262144,
+                "maxOutputTokens": 65536,
+                "pricingPerMTok": {
+                    "input": 0.3,
+                    "output": 2.0,
+                    "cachedInput": 0.03
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Qwen 3.6 27B — image input documented in input_modalities alongside the full feature set. credential-free roster probe: unauthenticated GET https://llm.chutes.ai/v1/models on 2026-09-28 (HTTP 200, 14 models). Roster-listed only; not capability-probed; fallback."
+            },
+            "Qwen/Qwen3.8-27B-TEE": {
+                "contextWindow": 262144,
+                "maxOutputTokens": 65536,
+                "pricingPerMTok": {
+                    "input": 0.24,
+                    "output": 2.2,
+                    "cachedInput": 0.024
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Qwen 3.8 27B — image input documented in input_modalities alongside the full feature set. credential-free roster probe: unauthenticated GET https://llm.chutes.ai/v1/models on 2026-09-28 (HTTP 200, 14 models). Roster-listed only; not capability-probed; fallback."
+            },
+            "Qwen/Qwen3-235B-A22B-Thinking-2507-TEE": {
+                "contextWindow": 262144,
+                "maxOutputTokens": 262144,
+                "pricingPerMTok": {
+                    "input": 0.2989,
+                    "output": 1.1957,
+                    "cachedInput": 0.0299
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Qwen 3 235B-A22B Thinking (2507 snapshot) — a reasoning-focused variant whose max_output_length equals its full 262,144-token context window, so a generous max_tokens budget matters most for this entry. Text-only input per input_modalities. credential-free roster probe: unauthenticated GET https://llm.chutes.ai/v1/models on 2026-09-28 (HTTP 200, 14 models). Roster-listed only; not capability-probed; fallback."
+            },
+            "deepseek-ai/DeepSeek-V3.2-TEE": {
+                "contextWindow": 131072,
+                "maxOutputTokens": 65536,
+                "pricingPerMTok": {
+                    "input": 1.0,
+                    "output": 1.0,
+                    "cachedInput": 0.1
+                },
+                "vision": false,
+                "status": "production",
+                "description": "DeepSeek V3.2 — text-only input per input_modalities, full documented feature set. credential-free roster probe: unauthenticated GET https://llm.chutes.ai/v1/models on 2026-09-28 (HTTP 200, 14 models). Roster-listed only; not capability-probed; fallback."
+            },
+            "zai-org/GLM-5.1-TEE": {
+                "contextWindow": 202752,
+                "maxOutputTokens": 65535,
+                "pricingPerMTok": {
+                    "input": 0.98,
+                    "output": 3.08,
+                    "cachedInput": 0.098
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Zhipu GLM-5.1 — prior GLM generation, text-only input per input_modalities. credential-free roster probe: unauthenticated GET https://llm.chutes.ai/v1/models on 2026-09-28 (HTTP 200, 14 models). Roster-listed only; not capability-probed; fallback."
+            },
+            "google/gemma-4-31B-turbo-TEE": {
+                "contextWindow": 131072,
+                "maxOutputTokens": 65536,
+                "pricingPerMTok": {
+                    "input": 0.12,
+                    "output": 0.37,
+                    "cachedInput": 0.012
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Google Gemma 4 31B Turbo — image input documented in input_modalities. The roster reports a 262,144 max_model_len against a smaller 131,072 context_length; contextWindow here uses the latter (context_length), the field name this catalog treats as the documented context window — see https://llm.chutes.ai/v1/models. credential-free roster probe: unauthenticated GET https://llm.chutes.ai/v1/models on 2026-09-28 (HTTP 200, 14 models). Roster-listed only; not capability-probed; fallback."
+            },
+            "Qwen/Qwen3-32B-TEE": {
+                "contextWindow": 40960,
+                "maxOutputTokens": 40960,
+                "pricingPerMTok": {
+                    "input": 0.104,
+                    "output": 0.416,
+                    "cachedInput": 0.0104
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Qwen 3 32B — smallest documented context window in the roster (40,960 tokens), text-only input. credential-free roster probe: unauthenticated GET https://llm.chutes.ai/v1/models on 2026-09-28 (HTTP 200, 14 models). Roster-listed only; not capability-probed; fallback."
+            },
+            "unsloth/Mistral-Nemo-Instruct-2407-TEE": {
+                "pricingPerMTok": {
+                    "input": 0.0245,
+                    "output": 0.0978,
+                    "cachedInput": 0.0024
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Mistral Nemo Instruct (2407, Unsloth build) — the public /v1/models entry gives only max_model_len (131,072); context_length, max_output_length, input_modalities and supported_features are all null/absent for this id, so no context window, output cap, vision or tool/JSON-mode capability is asserted here (contextWindow and maxOutputTokens are omitted rather than guessed from max_model_len, which is not documented as equivalent to either). credential-free roster probe: unauthenticated GET https://llm.chutes.ai/v1/models on 2026-09-28 (HTTP 200, 14 models). Cheapest model in the roster by list price. Roster-listed only; not capability-probed; fallback."
+            },
+            "Nemotron-3-Nano-Omni-30B-TEE": {
+                "pricingPerMTok": {
+                    "input": 0.0245,
+                    "output": 0.0978,
+                    "cachedInput": 0.0024
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Nemotron 3 Nano Omni 30B — same undocumented-field pattern as unsloth/Mistral-Nemo-Instruct-2407-TEE: only max_model_len (131,072) is given; context_length, max_output_length, input_modalities and supported_features are all null/absent. The \"Omni\" name suggests multimodal capability, but nothing in the public roster confirms it, so vision is left false here rather than inferred from the name. credential-free roster probe: unauthenticated GET https://llm.chutes.ai/v1/models on 2026-09-28 (HTTP 200, 14 models). Roster-listed only; not capability-probed; fallback."
+            }
+        },
+        "topModels": [
+            "moonshotai/Kimi-K2.6-TEE",
+            "moonshotai/Kimi-K3-TEE",
+            "deepseek-ai/DeepSeek-V4-Flash-0731-TEE",
+            "zai-org/GLM-5.2-TEE",
+            "Qwen/Qwen3.5-397B-A17B-TEE",
+            "deepseek-ai/DeepSeek-V3.2-TEE"
+        ]
+    },
+    "capabilities": {
+        "text": true,
+        "streaming": true,
+        "tools": "model-dependent",
+        "toolsWithStreaming": false,
+        "structuredOutput": true,
+        "structuredOutputWithTools": false,
+        "embeddings": false,
+        "thinking": true
+    },
+    "errorRules": [],
+    "setup": {
+        "url": "https://chutes.ai/signup",
+        "apiKeyFormat": "^cpk_[A-Za-z0-9_-]+$",
+        "billingPolicy": "no-free-tier",
+        "instructions": [
+            "1. Visit: https://chutes.ai/signup and create an account (pick a username; sign up with Google, GitHub, or optionally link a Bittensor wallet — no card is asked for on this page)",
+            "2. Manage billing and create API keys at https://api.chutes.ai (account/billing/API-key management is a separate host from the inference API)",
+            "3. Review current token pricing at https://chutes.ai/pricing before making requests — pay-as-you-go per token (USD per 1M tokens) plus optional paid Plus/Pro plans; no free trial credits are documented, so budget for a funded account from the first call",
+            "4. Set {apiKeyEnvVar} in your .env file — keys are issued with a cpk_ prefix and must be sent as Authorization: Bearer {apiKeyEnvVar} (X-API-Key is not a supported auth scheme)"
+        ],
+        "description": "OpenAI-compatible endpoint at https://llm.chutes.ai/v1 (chat completions confirmed via the vendor's own homepage code sample), serving a small TEE (confidential-compute) model roster — 14 ids at the time of writing, confirmed by an unauthenticated GET /v1/models. Billing is pay-as-you-go per token; the pricing page's free-trial FAQ answer is collapsed/unconfirmed and no free-tier language appears elsewhere in the public docs, so billingPolicy is recorded conservatively as no-free-tier pending a live-key check."
+    },
+    "evidence": {
+        "rosterVerified": {
+            "date": "2026-09-28",
+            "method": "unauthenticated GET https://llm.chutes.ai/v1/models — 14 models; credential-free, no key used",
+            "status": 200
+        },
+        "liveMatrix": null,
+        "addedInPR": "https://github.com/juspay/neurolink/pull/1848"
+    }
+}

package/dist/providers/catalog/deepinfra.json

@@ -0,0 +1,157 @@
+{
+    "$schema": "./provider-catalog.schema.json",
+    "id": "deepinfra",
+    "displayName": "DeepInfra",
+    "aliases": [],
+    "tier": 2,
+    "wire": {
+        "baseURL": "https://api.deepinfra.com/v1/openai"
+    },
+    "models": {
+        "default": "deepseek-ai/DeepSeek-V4-Flash-0731",
+        "fallbacks": [
+            "deepseek-ai/DeepSeek-V4-Pro-0813",
+            "openai/gpt-oss-120b",
+            "moonshotai/Kimi-K3",
+            "zai-org/GLM-5.2",
+            "anthropic/claude-haiku-4-5",
+            "Qwen/Qwen3-14B",
+            "meta-llama/Llama-4-Scout-17B-16E-Instruct"
+        ],
+        "defaultContextWindow": 128000,
+        "defaultMaxOutputTokens": 16384,
+        "catalog": {
+            "deepseek-ai/DeepSeek-V4-Flash-0731": {
+                "contextWindow": 1048576,
+                "pricingPerMTok": {
+                    "input": 0.06,
+                    "output": 0.18,
+                    "cachedInput": 0.015
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Recommended default — DeepInfra's own tool-calling documentation (docs.deepinfra.com/chat/tool-calling, retrieved 2026-09-28) uses this exact model in its function-calling example (a get_current_weather tool_call round-trip), and its reasoning documentation (docs.deepinfra.com/chat/reasoning) names it among the models that support the reasoning_effort parameter (none/low/medium/high). Live on the unauthenticated GET /v1/openai/models roster (2026-09-28, 187 models) with context_length/max_tokens both 1,048,576 and pricing $0.06/$0.18 per MTok input/output ($0.015 cached-input read). No vision/vlm tag on the live roster. models.defaultMaxOutputTokens (16,384) follows the vendor's documented general cap rather than this per-model figure — see models-level note."
+            },
+            "deepseek-ai/DeepSeek-V4-Pro-0813": {
+                "contextWindow": 1048576,
+                "pricingPerMTok": {
+                    "input": 1.3,
+                    "output": 2.6,
+                    "cachedInput": 0.1
+                },
+                "vision": false,
+                "status": "production",
+                "description": "DeepSeek's larger Pro variant; roster-listed only (not itself named in a DeepInfra docs example) via the unauthenticated GET /v1/openai/models (2026-09-28) — context 1,048,576, $1.30/$2.60 per MTok input/output, $0.10 cached-input read. Fallback."
+            },
+            "openai/gpt-oss-120b": {
+                "contextWindow": 131072,
+                "pricingPerMTok": {
+                    "input": 0.037,
+                    "output": 0.17
+                },
+                "vision": false,
+                "status": "production",
+                "description": "OpenAI's open-weight 120B model hosted on DeepInfra; tagged reasoning + reasoning_effort on the live roster GET (2026-09-28) — context 131,072, $0.037/$0.17 per MTok input/output. Not itself named in DeepInfra's docs examples; fallback."
+            },
+            "moonshotai/Kimi-K3": {
+                "contextWindow": 1048576,
+                "pricingPerMTok": {
+                    "input": 2.85,
+                    "output": 14.25,
+                    "cachedInput": 0.285
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Named in DeepInfra's tool-calling documentation as \"the current Kimi model in our catalog\" (docs.deepinfra.com/chat/tool-calling). Tagged vlm + vision on the live roster GET (2026-09-28), hence vision: true here (not independently probed for a working vision reply) — context 1,048,576, $2.85/$14.25 per MTok input/output, $0.285 cached-input read. Fallback."
+            },
+            "zai-org/GLM-5.2": {
+                "contextWindow": 1048576,
+                "pricingPerMTok": {
+                    "input": 0.75,
+                    "output": 2.4,
+                    "cachedInput": 0.14
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Listed by DeepInfra's reasoning documentation (docs.deepinfra.com/chat/reasoning) as one of the models supporting reasoning_effort. Live roster GET (2026-09-28) — context 1,048,576, $0.75/$2.40 per MTok input/output, $0.14 cached-input read. Fallback."
+            },
+            "anthropic/claude-haiku-4-5": {
+                "contextWindow": 200000,
+                "pricingPerMTok": {
+                    "input": 1.0,
+                    "output": 5.0
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Anthropic's Claude Haiku 4.5 as served through DeepInfra's model marketplace (owned_by: deepinfra on the live roster). Tagged vlm + vision + reasoning_effort + reasoning on the GET /v1/openai/models roster (2026-09-28), hence vision: true here (not independently probed) — context 200,000, $1.00/$5.00 per MTok input/output. Not itself named in DeepInfra's docs examples; fallback."
+            },
+            "Qwen/Qwen3-14B": {
+                "contextWindow": 40960,
+                "pricingPerMTok": {
+                    "input": 0.12,
+                    "output": 0.24
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Smaller Qwen reasoning-tagged model; roster-listed only via the unauthenticated GET /v1/openai/models (2026-09-28) — context 40,960, $0.12/$0.24 per MTok input/output. Fallback."
+            },
+            "meta-llama/Llama-4-Scout-17B-16E-Instruct": {
+                "contextWindow": 327680,
+                "pricingPerMTok": {
+                    "input": 0.1,
+                    "output": 0.3
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Meta Llama 4 Scout; tagged vlm + vision on the live roster GET (2026-09-28), hence vision: true here (not independently probed) — context 327,680, $0.10/$0.30 per MTok input/output. Roster-listed only; fallback."
+            }
+        },
+        "topModels": [
+            "deepseek-ai/DeepSeek-V4-Flash-0731",
+            "deepseek-ai/DeepSeek-V4-Pro-0813",
+            "openai/gpt-oss-120b",
+            "moonshotai/Kimi-K3",
+            "zai-org/GLM-5.2",
+            "anthropic/claude-haiku-4-5",
+            "Qwen/Qwen3-14B",
+            "meta-llama/Llama-4-Scout-17B-16E-Instruct"
+        ],
+        "visionModel": "meta-llama/Llama-4-Scout-17B-16E-Instruct"
+    },
+    "capabilities": {
+        "text": true,
+        "streaming": true,
+        "tools": "model-dependent",
+        "toolsWithStreaming": true,
+        "structuredOutput": true,
+        "structuredOutputWithTools": false,
+        "embeddings": false,
+        "thinking": false
+    },
+    "errorRules": [],
+    "setup": {
+        "url": "https://deepinfra.com/dash/api_keys",
+        "apiKeyFormat": null,
+        "billingPolicy": "no-free-tier",
+        "instructions": [
+            "1. Visit: https://deepinfra.com/dash/api_keys and sign in (Google, GitHub, email & password, or Corporate SSO — see https://docs.deepinfra.com/account/signing-in)",
+            "2. Create an API key and copy it when it is shown",
+            "3. No free tier: DeepInfra requires a card or pre-payment before any request succeeds (https://deepinfra.com/pricing)",
+            "4. Set {apiKeyEnvVar} in your .env file"
+        ],
+        "description": "OpenAI-compatible endpoint at https://api.deepinfra.com/v1/openai (docs.deepinfra.com/chat/overview), serving 180+ open-weight and proprietary chat models. No documented auth-failure error body was found on the docs site or the public OpenAPI spec (api.deepinfra.com/openapi.json only documents 200/422 for chat/completions) without a live credentialed probe, so errorRules is empty pending live verification."
+    },
+    "evidence": {
+        "rosterVerified": {
+            "date": "2026-09-28",
+            "method": "unauthenticated GET https://api.deepinfra.com/v1/openai/models — 187 models; credential-free, no key used",
+            "status": 200
+        },
+        "billingProbe": {
+            "date": "2026-09-28",
+            "method": "public page https://deepinfra.com/pricing — \"You have to add a card or pre-pay or you won't be able to use our services\"; no free tier documented; credential-free"
+        },
+        "liveMatrix": null,
+        "addedInPR": "https://github.com/juspay/neurolink/pull/1848"
+    }
+}

package/dist/providers/catalog/empiriolabs.json

@@ -0,0 +1,143 @@
+{
+    "$schema": "./provider-catalog.schema.json",
+    "id": "empiriolabs",
+    "displayName": "EmpirioLabs AI",
+    "aliases": [],
+    "tier": 2,
+    "wire": {
+        "baseURL": "https://api.empiriolabs.ai/v1"
+    },
+    "models": {
+        "default": "glm-5-3",
+        "fallbacks": [
+            "glm-5-3-flash",
+            "glm-5-2",
+            "kimi-k3",
+            "qwen3-7-max",
+            "minimax-m2-7-highspeed"
+        ],
+        "defaultContextWindow": 1000000,
+        "defaultMaxOutputTokens": 131072,
+        "catalog": {
+            "glm-5-3": {
+                "contextWindow": 1000000,
+                "maxOutputTokens": 131072,
+                "pricingPerMTok": {
+                    "input": 1.4,
+                    "output": 4.4,
+                    "cachedInput": 1.4
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Recommended — EmpirioLabs' own roster marks this is_featured=true (GET /v1/models, 2026-09-28). Zhipu GLM-5.3: reasoning model with native function calling and web search, response_format json_object structured output, 1M token context, 131,072 token max output — all read directly from the unauthenticated GET https://api.empiriolabs.ai/v1/models. Pricing from the same roster: $1.40 / $4.40 per MTok input/output, $1.40 per MTok cached input. Onboarded credential-free — no live chat, streaming or tool-call request was made, so wire behavior (SSE framing, a real tool_call round-trip, image input) is roster- and docs-verified only, not live-probed; see evidence.liveMatrix (null)."
+            },
+            "glm-5-3-flash": {
+                "contextWindow": 1000000,
+                "maxOutputTokens": 131072,
+                "pricingPerMTok": {
+                    "input": 0.075,
+                    "output": 0.25,
+                    "cachedInput": 0.075
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Zhipu GLM-5.3 Flash — lighter/cheaper sibling of the default. The roster marks capabilities.images=true (also video_understanding and document_understanding features), function calling and json_object structured output. 1M context, 131,072 max output. Pricing: $0.075 / $0.25 per MTok input/output, $0.075 cached. Roster-listed only, not capability-probed (credential-free onboarding — no live request made); fallback."
+            },
+            "glm-5-2": {
+                "contextWindow": 1000000,
+                "maxOutputTokens": 131072,
+                "pricingPerMTok": {
+                    "input": 1.4,
+                    "output": 4.4,
+                    "cachedInput": 1.4
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Zhipu GLM-5.2 — prior GLM generation; reasoning, function calling, web search; json_object structured output; 1M context, 131,072 max output. Pricing $1.40 / $4.40 per MTok. Roster-listed only, not capability-probed; fallback."
+            },
+            "kimi-k3": {
+                "contextWindow": 1000000,
+                "maxOutputTokens": 131072,
+                "pricingPerMTok": {
+                    "input": 3,
+                    "output": 15,
+                    "cachedInput": 3
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Moonshot Kimi K3 — reasoning, function calling, multimodal (capabilities.images=true on the roster), agentic coding and web search; strict json_schema structured output (a stronger guarantee than glm-5-3's json_object). 1M context, 131,072 max output. Pricing $3 / $15 per MTok. Roster-listed only, not capability-probed; fallback."
+            },
+            "qwen3-7-max": {
+                "contextWindow": 1000000,
+                "maxOutputTokens": 65536,
+                "pricingPerMTok": {
+                    "input": 2.5,
+                    "output": 7.5,
+                    "cachedInput": 2.5
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Alibaba Qwen3.7 Max — flagship text reasoning model with code_interpreter, function calling and agentic coding; strict json_schema structured output. 1M context, 65,536 max output (lower than the GLM family's 131,072 — read from the roster, not assumed). Pricing $2.50 / $7.50 per MTok. Roster-listed only, not capability-probed; fallback."
+            },
+            "minimax-m2-7-highspeed": {
+                "contextWindow": 200000,
+                "maxOutputTokens": 32768,
+                "pricingPerMTok": {
+                    "input": 0.3,
+                    "output": 1.2,
+                    "cachedInput": 0.03
+                },
+                "vision": false,
+                "status": "production",
+                "description": "MiniMax M2.7 Highspeed — reasoning and function calling only (no vision or web_search declared on the roster); json_object structured output. Smallest context in this curated set (200K, 32,768 max output) — kept as the lightest/cheapest fallback. Pricing $0.30 / $1.20 per MTok, $0.03 cached. Roster-listed only, not capability-probed; fallback."
+            }
+        },
+        "topModels": [
+            "glm-5-3",
+            "glm-5-3-flash",
+            "glm-5-2",
+            "kimi-k3",
+            "qwen3-7-max",
+            "minimax-m2-7-highspeed"
+        ],
+        "visionModel": "glm-5-3-flash"
+    },
+    "capabilities": {
+        "text": true,
+        "streaming": true,
+        "tools": "model-dependent",
+        "toolsWithStreaming": false,
+        "structuredOutput": true,
+        "structuredOutputWithTools": false,
+        "embeddings": false,
+        "thinking": false
+    },
+    "errorRules": [
+        {
+            "status": 401,
+            "class": "authentication",
+            "message": "Invalid EmpirioLabs API key. Check {apiKeyEnvVar}. Get one at {setupUrl}"
+        }
+    ],
+    "setup": {
+        "url": "https://platform.empiriolabs.ai",
+        "apiKeyFormat": "^sk-empiriolabs-",
+        "billingPolicy": "free-tier",
+        "instructions": [
+            "1. Visit: https://platform.empiriolabs.ai and create an account",
+            "2. Create an API key from the dashboard (keys use the sk-empiriolabs- prefix, up to 50 per account)",
+            "3. Selected models (e.g. GLM 4.7 Flash) run at $0 cost; other models draw from prepaid credits topped up via the dashboard Billing page — confirm current pricing before heavy use",
+            "4. Set {apiKeyEnvVar} in your .env file"
+        ],
+        "description": "OpenAI-compatible endpoint at https://api.empiriolabs.ai/v1 (docs.empiriolabs.ai confirms it accepts the OpenAI Chat Completions request format). Hosts a large catalog — 217 models on an unauthenticated GET /v1/models as of 2026-09-28 — spanning text, vision, audio, video, image and 3D generation; this catalog entry covers only a curated set of text/chat-completions-capable models. New accounts default to 50 RPM / 2,000,000 TPM."
+    },
+    "evidence": {
+        "rosterVerified": {
+            "date": "2026-09-28",
+            "method": "unauthenticated GET https://api.empiriolabs.ai/v1/models — 217 models; credential-free, no key used",
+            "status": 200
+        },
+        "liveMatrix": null,
+        "addedInPR": "https://github.com/juspay/neurolink/pull/1848"
+    }
+}