@juspay/neurolink 12.39.1 → 12.40.0

package/dist/providers/catalog/moonshot-ai.json

@@ -0,0 +1,135 @@
+{
+    "$schema": "./provider-catalog.schema.json",
+    "id": "moonshot-ai",
+    "displayName": "Moonshot AI (Kimi)",
+    "aliases": ["moonshot", "kimi"],
+    "tier": 2,
+    "wire": {
+        "baseURL": "https://api.moonshot.ai/v1",
+        "apiKeyFallbackEnvVars": ["MOONSHOT_API_KEY"]
+    },
+    "models": {
+        "default": "kimi-k3",
+        "fallbacks": ["kimi-k2.7-code", "kimi-k2.6", "kimi-k2.7-code-highspeed"],
+        "defaultContextWindow": 262144,
+        "defaultMaxOutputTokens": 131072,
+        "catalog": {
+            "kimi-k3": {
+                "contextWindow": 1048576,
+                "pricingPerMTok": {
+                    "input": 3.0,
+                    "output": 15.0,
+                    "cachedInput": 0.3
+                },
+                "vision": true,
+                "status": "production",
+                "description": "NeuroLink default. Model list https://platform.kimi.ai/docs/models (retrieved 2026-09-29): \"Kimi's most capable model to date, with 2.8 trillion parameters, native visual understanding, and a 1M-token context window, designed for frontier intelligence scenarios such as software engineering, knowledge work, and deep reasoning.\" The Kimi K3 page https://platform.kimi.ai/docs/guide/kimi-k3-quickstart calls it \"Kimi's most capable flagship model to date\", states \"K3 always has thinking mode enabled\" with the top-level reasoning_effort field (low, high, max; default max), and states max_completion_tokens \"defaults to 131072 and can be set up to 1048576\". Pricing page https://platform.kimi.ai/docs/pricing/chat (2026-09-29): context window 1,048,576 tokens; per 1M tokens input $3.00, cached input $0.30, output $15.00, cache write $3.00 (TTL 5min) or $6.00 (TTL 1h). The vision guide https://platform.kimi.ai/docs/guide/use-kimi-vision-model names kimi-k3 in its description of the Kimi Vision Model, hence vision true."
+            },
+            "kimi-k2.7-code": {
+                "contextWindow": 262144,
+                "pricingPerMTok": {
+                    "input": 0.95,
+                    "output": 4.0,
+                    "cachedInput": 0.19
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Fallback in NeuroLink's catalog order. Model list https://platform.kimi.ai/docs/models (retrieved 2026-09-29): \"Kimi's dedicated coding model.\" with \"Context 256k\". Pricing page https://platform.kimi.ai/docs/pricing/chat (2026-09-29): context window 262,144 tokens; per 1M tokens cache-hit input $0.19, cache-miss input $0.95, output $4.00. The Kimi K2.7 Code page https://platform.kimi.ai/docs/guide/kimi-k2-7-code-quickstart states \"Kimi K2.7 Code does not support non-thinking mode.\" The parameter reference https://platform.kimi.ai/docs/api/models-overview states that kimi-k2.6 and kimi-k2.7-code do not support tool_choice required and \"return an error if it is passed\". The vision guide https://platform.kimi.ai/docs/guide/use-kimi-vision-model names kimi-k2.7-code in its description of the Kimi Vision Model, hence vision true."
+            },
+            "kimi-k2.7-code-highspeed": {
+                "contextWindow": 262144,
+                "pricingPerMTok": {
+                    "input": 1.9,
+                    "output": 8.0,
+                    "cachedInput": 0.38
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Fallback in NeuroLink's catalog order. Model list https://platform.kimi.ai/docs/models (retrieved 2026-09-29): \"High-Speed version of Kimi K2.7 Code model, with output speed of approximately 180 Tokens/s and up to 260 Tokens/s in short context scenarios\". The Kimi K2.7 Code page https://platform.kimi.ai/docs/guide/kimi-k2-7-code-quickstart states it is \"the same model as Kimi K2.7 Code\" and ends that paragraph with \"(Currently, the resource is limited, and the experience of the high-speed model may be slightly fluctuate，we are gradually increasing the resource.)\". Pricing page https://platform.kimi.ai/docs/pricing/chat (2026-09-29): context window 262,144 tokens; per 1M tokens cache-hit input $0.38, cache-miss input $1.90, output $8.00. The vision guide https://platform.kimi.ai/docs/guide/use-kimi-vision-model names kimi-k2.7-code-highspeed in its description of the Kimi Vision Model, hence vision true."
+            },
+            "kimi-k2.6": {
+                "contextWindow": 262144,
+                "pricingPerMTok": {
+                    "input": 0.95,
+                    "output": 4.0,
+                    "cachedInput": 0.16
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Fallback in NeuroLink's catalog order. Model list https://platform.kimi.ai/docs/models (retrieved 2026-09-29): \"Supports both visual and text input, thinking and non-thinking modes, and dialogue and Agent tasks.\" with \"Context 256k\". Pricing page https://platform.kimi.ai/docs/pricing/chat (2026-09-29): context window 262,144 tokens; per 1M tokens cache-hit input $0.16, cache-miss input $0.95, output $4.00. The Thinking Models page https://platform.kimi.ai/docs/guide/use-thinking-models states \"thinking is on by default, can be disabled\" for kimi-k2.6. The response_format guide https://platform.kimi.ai/docs/guide/response_format states \"kimi-k2.6 occasionally behaves unstably with complex schemas; for example, $ref may return a Markdown code block, oneOf may be ignored, and partial=true may output fields outside the schema.\" The vision guide https://platform.kimi.ai/docs/guide/use-kimi-vision-model names kimi-k2.6 in its description of the Kimi Vision Model, hence vision true."
+            }
+        },
+        "topModels": [
+            "kimi-k3",
+            "kimi-k2.7-code",
+            "kimi-k2.7-code-highspeed",
+            "kimi-k2.6"
+        ]
+    },
+    "capabilities": {
+        "text": true,
+        "streaming": true,
+        "tools": true,
+        "toolsWithStreaming": true,
+        "structuredOutput": true,
+        "structuredOutputWithTools": false,
+        "embeddings": false,
+        "thinking": false
+    },
+    "errorRules": [
+        {
+            "pattern": "Account balance is insufficient|Token quota is insufficient|exceeded_current_quota_error",
+            "class": "provider",
+            "message": "Moonshot AI reports an insufficient account balance or token quota (HTTP 429, exceeded_current_quota_error). Its error list gives the messages \"Account balance is insufficient or the account has been disabled\" and \"Token quota is insufficient\" (https://platform.kimi.ai/docs/api/errors)"
+        },
+        {
+            "status": 401,
+            "class": "authentication",
+            "message": "Invalid or missing Moonshot AI API key (HTTP 401, incorrect_api_key_error: \"The API key was not provided, or the key is incorrect\"). Check {apiKeyEnvVar}. Get one at {setupUrl}. Keys issued on platform.kimi.ai are independent from keys issued on other regional Kimi platforms (https://platform.kimi.ai/docs/api/errors)"
+        },
+        {
+            "status": 404,
+            "class": "invalid-model",
+            "message": "Model {model} was not found, or this account does not have permission to access it (HTTP 404, resource_not_found_error: \"Model not found, or this account does not have permission to access the model\"; https://platform.kimi.ai/docs/api/errors)"
+        },
+        {
+            "status": 429,
+            "class": "rate-limit",
+            "message": "Moonshot AI returned HTTP 429. Its error list gives engine_overloaded_error, exceeded_current_quota_error and rate_limit_reached_error for this status (https://platform.kimi.ai/docs/api/errors)"
+        }
+    ],
+    "quirks": {
+        "replayReasoningContent": true
+    },
+    "setup": {
+        "url": "https://platform.kimi.ai/console/api-keys",
+        "apiKeyFormat": null,
+        "billingPolicy": "no-free-tier",
+        "instructions": [
+            "1. Visit: https://platform.kimi.ai/console/api-keys and sign in, then create and copy an API key (https://platform.kimi.ai/docs/overview describes signing in to the Kimi API Platform and creating a key under API Keys)",
+            "2. Create the key on platform.kimi.ai: the errors page states that keys issued on platform.kimi.ai are independent from keys issued on other regional Kimi platforms, and that \"Mixing keys across platforms returns 401.\" (https://platform.kimi.ai/docs/api/errors)",
+            "3. Top up before use: \"To prevent abuse, you need to recharge at least $1 to start using, and when your cumulative recharge reaches $5, you will receive a $5 voucher.\" (https://platform.kimi.ai/docs/pricing/limits); the Kimi K3 page states \"Kimi K3 is a flagship model: it is unlocked after a successful top-up (minimum $1).\" (https://platform.kimi.ai/docs/guide/kimi-k3-quickstart)",
+            "4. Set {apiKeyEnvVar} in your .env file (MOONSHOT_API_KEY, the variable Kimi's own docs use, is also read)"
+        ],
+        "description": "OpenAI-compatible endpoint at https://api.moonshot.ai/v1 (https://platform.kimi.ai/docs/overview: \"Kimi API lets you interact with Kimi models and is compatible with both the OpenAI and Anthropic API formats.\"). The docs and console live on platform.kimi.ai and call the product the Kimi API; the API host is api.moonshot.ai. This entry is docs-verified only: the vendor's GET /v1/models needs a key, so the four model ids come from its public docs and the roster has not been checked; it is not yet live-verified. models.defaultContextWindow (262144, the context window the pricing page lists for kimi-k2.7-code, kimi-k2.7-code-highspeed and kimi-k2.6) and models.defaultMaxOutputTokens (131072, the max_completion_tokens default the vendor states for kimi-k3) apply to model ids outside the catalog and are placeholders, since the vendor publishes no figure for such ids. errorRules follow the statuses and messages on https://platform.kimi.ai/docs/api/errors; the 401 status and incorrect_api_key_error type were also seen on an unauthenticated GET of /v1/models."
+    },
+    "evidence": {
+        "rosterVerified": {
+            "date": "2026-09-29",
+            "status": 401,
+            "method": "docs-only — unauthenticated GET https://api.moonshot.ai/v1/models answers 401 without a key, so model ids are taken from https://platform.kimi.ai/docs/models (retrieved 2026-09-29); roster not verified"
+        },
+        "authProbe": {
+            "date": "2026-09-29",
+            "status": 401,
+            "code": "incorrect_api_key_error",
+            "method": "unauthenticated GET https://api.moonshot.ai/v1/models with no credentials — HTTP 401, JSON body {\"error\":{\"message\":\"Incorrect API key provided\",\"type\":\"incorrect_api_key_error\"}}; the code recorded here is the body's error.type value (the body has no error.code field); credential-free, no key used"
+        },
+        "billingProbe": {
+            "date": "2026-09-29",
+            "method": "public page https://platform.kimi.ai/docs/pricing/limits — \"To prevent abuse, you need to recharge at least $1 to start using, and when your cumulative recharge reaches $5, you will receive a $5 voucher.\"; credential-free"
+        },
+        "liveMatrix": null,
+        "addedInPR": "https://github.com/juspay/neurolink/pull/1860"
+    }
+}

package/dist/providers/catalog/nebius.json

@@ -0,0 +1,165 @@
+{
+    "$schema": "./provider-catalog.schema.json",
+    "id": "nebius",
+    "displayName": "Nebius Token Factory",
+    "aliases": [],
+    "tier": 2,
+    "wire": {
+        "baseURL": "https://api.tokenfactory.nebius.com/v1"
+    },
+    "models": {
+        "default": "Qwen/Qwen3-235B-A22B-Instruct-2507",
+        "fallbacks": [
+            "deepseek-ai/DeepSeek-V4-Flash-0731",
+            "openai/gpt-oss-120b",
+            "Qwen/Qwen3-30B-A3B-Instruct-2507",
+            "NousResearch/Hermes-4-405B"
+        ],
+        "defaultContextWindow": 32000,
+        "defaultMaxOutputTokens": 8192,
+        "catalog": {
+            "Qwen/Qwen3-235B-A22B-Instruct-2507": {
+                "contextWindow": 262144,
+                "pricingPerMTok": {
+                    "input": 0.2,
+                    "output": 0.6
+                },
+                "vision": false,
+                "status": "production",
+                "description": "NeuroLink default. Public model catalog (https://tokenfactory.nebius.com/api/public/models_info, retrieved 2026-09-29): name \"Qwen3-235B-A22B-Instruct-2507\"; type text2text; status active; description \"Balanced Qwen3 flagship tuned for strong general reasoning, chat quality, and tool use.\"; tags \"262K context\", \"code\", \"math\", \"JSON mode\"; use_cases text, code, function_calling, context_and_rag, fine_tune, responses_api; context_window_k 262; max_model_len 262144; input_price_per_million_tokens 0.2; output_price_per_million_tokens 0.6. vision false: catalog type text2text."
+            },
+            "deepseek-ai/DeepSeek-V4-Flash-0731": {
+                "contextWindow": 1024000,
+                "pricingPerMTok": {
+                    "input": 0.14,
+                    "output": 0.28
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Fallback. Public model catalog (https://tokenfactory.nebius.com/api/public/models_info, retrieved 2026-09-29): name \"DeepSeek-V4-Flash-0731\"; type text2text; status active; description \"DeepSeek V4 Flash 0731 is a 1M-context reasoning model designed for coding and agentic workloads.\"; tags \"1M context\", \"reasoning\", \"JSON mode\", \"code\", \"math\"; use_cases text, reasoning, function_calling, code, context_and_rag, responses_api, fine_tune; context_window_k 1024; max_model_len 1024000; input_price_per_million_tokens 0.14; output_price_per_million_tokens 0.28. vision false: catalog type text2text. The August 2026 deprecation notice (https://docs.tokenfactory.nebius.com/august-2026-deprecation-notice) lists this id in its \"Recommended Replacement\" column."
+            },
+            "openai/gpt-oss-120b": {
+                "contextWindow": 131072,
+                "pricingPerMTok": {
+                    "input": 0.15,
+                    "output": 0.6
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Fallback. Public model catalog (https://tokenfactory.nebius.com/api/public/models_info, retrieved 2026-09-29): name \"gpt-oss-120b\"; type text2text; status active; description \"Open-weight agentic model with configurable reasoning, full CoT visibility, strong tool use, and fine-tuning support.\"; tags \"131K context\", \"code\", \"JSON mode\", \"math\", \"reasoning\"; use_cases text, code, fine_tune, function_calling, reasoning; context_window_k 131; max_model_len 131072; input_price_per_million_tokens 0.15; output_price_per_million_tokens 0.6. vision false: catalog type text2text. The Structured output & JSON page (https://docs.tokenfactory.nebius.com/ai-models-inference/json) uses this id in its JavaScript example."
+            },
+            "Qwen/Qwen3-30B-A3B-Instruct-2507": {
+                "contextWindow": 262144,
+                "pricingPerMTok": {
+                    "input": 0.1,
+                    "output": 0.3
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Fallback. Public model catalog (https://tokenfactory.nebius.com/api/public/models_info, retrieved 2026-09-29): name \"Qwen3-30B-A3B-Instruct-2507\"; type text2text; status active; description \"Versatile 30B instruct model optimized for high-quality chat, reasoning, and coding.\"; tags \"262K context\", \"code\", \"math\", \"JSON mode\"; use_cases text, code, function_calling, context_and_rag, fine_tune, responses_api; context_window_k 262; max_model_len 262144; input_price_per_million_tokens 0.1; output_price_per_million_tokens 0.3. vision false: catalog type text2text."
+            },
+            "NousResearch/Hermes-4-405B": {
+                "contextWindow": 131072,
+                "pricingPerMTok": {
+                    "input": 1,
+                    "output": 3
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Fallback. Public model catalog (https://tokenfactory.nebius.com/api/public/models_info, retrieved 2026-09-29): name \"Hermes-4-405B\"; type text2text; status active; description \"Hybrid-reasoning model trained on verified CoT traces for strong math, coding, and step-by-step reliability.\"; tags \"128K context\", \"JSON mode\"; use_cases text, code, complex_writing, context_and_rag, function_calling, reasoning, summarization, responses_api; context_window_k 128; max_model_len 131072; input_price_per_million_tokens 1; output_price_per_million_tokens 3. vision false: catalog type text2text."
+            },
+            "nvidia/Nemotron-3_5-Lightning": {
+                "contextWindow": 1048576,
+                "pricingPerMTok": {
+                    "input": 0.06,
+                    "output": 0.24
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Public model catalog (https://tokenfactory.nebius.com/api/public/models_info, retrieved 2026-09-29): name \"Nemotron-3.5-Lightning\"; type text2text; status active; description \"NVIDIA's 30B-parameter hybrid MoE model with 3B active parameters per token, designed for efficient agentic reasoning, tool use, coding, and long-context workflows.\"; tags \"1M context\", \"reasoning\", \"code\", \"function calling\", \"MTP\"; use_cases text, reasoning, function_calling, code, context_and_rag, responses_api; context_window_k 1024; max_model_len 1048576; input_price_per_million_tokens 0.06; output_price_per_million_tokens 0.24. vision false: catalog type text2text. The August 2026 deprecation notice (https://docs.tokenfactory.nebius.com/august-2026-deprecation-notice) lists this id in its \"Recommended Replacement\" column."
+            },
+            "openbmb/MiniCPM-V-4_5": {
+                "contextWindow": 32000,
+                "pricingPerMTok": {
+                    "input": 0.658,
+                    "output": 1.11
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Public model catalog (https://tokenfactory.nebius.com/api/public/models_info, retrieved 2026-09-29): name \"openbmb/MiniCPM-V-4_5\"; type image2text; status active; description \"MiniCPM-V-4.5 – Compact multimodal model for image, multi-image, high-FPS/long-video, OCR/PDF understanding, with switchable fast/deep thinking.\"; tags \"JSON mode\"; use_cases text, image, summarization, context_and_rag; context_window_k 32; max_model_len 32000; input_price_per_million_tokens 0.658; output_price_per_million_tokens 1.11. vision true: catalog type image2text with image in use_cases."
+            }
+        },
+        "topModels": [
+            "Qwen/Qwen3-235B-A22B-Instruct-2507",
+            "deepseek-ai/DeepSeek-V4-Flash-0731",
+            "openai/gpt-oss-120b",
+            "Qwen/Qwen3-30B-A3B-Instruct-2507",
+            "NousResearch/Hermes-4-405B",
+            "nvidia/Nemotron-3_5-Lightning",
+            "openbmb/MiniCPM-V-4_5"
+        ],
+        "visionModel": "openbmb/MiniCPM-V-4_5"
+    },
+    "capabilities": {
+        "text": true,
+        "streaming": true,
+        "tools": "model-dependent",
+        "toolsWithStreaming": true,
+        "structuredOutput": true,
+        "structuredOutputWithTools": false,
+        "embeddings": false,
+        "thinking": false
+    },
+    "errorRules": [
+        {
+            "status": 401,
+            "class": "authentication",
+            "message": "Nebius Token Factory returned HTTP 401. Check {apiKeyEnvVar}. Get one at {setupUrl}"
+        },
+        {
+            "status": 402,
+            "class": "provider",
+            "message": "Nebius Token Factory returned HTTP 402 (Payment Required): the account budget is exhausted. Billing details: https://docs.tokenfactory.nebius.com/other-capabilities/billing-new"
+        },
+        {
+            "status": 404,
+            "class": "invalid-model",
+            "message": "Nebius Token Factory returned HTTP 404: the model '{model}' does not exist. Model ids are listed at https://tokenfactory.nebius.com/model-catalog.md"
+        },
+        {
+            "status": 429,
+            "class": "rate-limit",
+            "message": "Nebius Token Factory returned HTTP 429 (rate limit exceeded). The Retry-After response header gives the time in seconds to wait before making another request (https://docs.tokenfactory.nebius.com/ai-models-inference/rate-limits)"
+        }
+    ],
+    "setup": {
+        "url": "https://tokenfactory.nebius.com/",
+        "apiKeyFormat": null,
+        "billingPolicy": "free-with-card",
+        "instructions": [
+            "1. Visit: https://tokenfactory.nebius.com/ and create an account; the quickstart (https://docs.tokenfactory.nebius.com/quickstart) says: \"Log in using your Google or GitHub account\"",
+            "2. Create an API key (https://docs.tokenfactory.nebius.com/api-reference/introduction#authentication): go to the API keys section, click Create API key, enter the key name, click Create, and save the displayed key — \"Save the displayed API key. You cannot open it later in Nebius Token Factory.\"",
+            "3. Billing (https://docs.tokenfactory.nebius.com/other-capabilities/billing-new): \"Billing setup is mandatory—you cannot complete onboarding without it.\" \"Setting up a billing account requires a bank card.\" \"Upon first sign-up, you receive $1 in trial credit, valid for 30 days.\"",
+            "4. Set {apiKeyEnvVar} in your .env file (the vendor's docs read NEBIUS_API_KEY)"
+        ],
+        "description": "OpenAI-compatible endpoint at https://api.tokenfactory.nebius.com/v1 (https://docs.tokenfactory.nebius.com/api-reference/introduction: \"Nebius Token Factory offers an OpenAI-compatible API for inference and fine-tuning.\"). Docs-verified, not live-verified: an unauthenticated GET of https://api.tokenfactory.nebius.com/v1/models answers HTTP 401 without a key, so the model ids come from the vendor's public model catalog (https://tokenfactory.nebius.com/api/public/models_info, retrieved 2026-09-29) and the roster has not been checked against the API. errorRules follow the statuses and example bodies in https://docs.tokenfactory.nebius.com/api-reference/inference/create-chat-completion (401, 402, 404, 429). models.defaultMaxOutputTokens (8192) is the value that reference gives for an omitted max_tokens (\"If omitted or set to null, defaults to 8192 tokens.\"); models.defaultContextWindow (32000) is a placeholder the vendor does not publish."
+    },
+    "evidence": {
+        "rosterVerified": {
+            "date": "2026-09-29",
+            "status": 401,
+            "method": "docs-only — unauthenticated GET https://api.tokenfactory.nebius.com/v1/models answers 401 without a key, so model ids are taken from https://tokenfactory.nebius.com/api/public/models_info (retrieved 2026-09-29); roster not verified"
+        },
+        "authProbe": {
+            "date": "2026-09-29",
+            "status": 401,
+            "method": "unauthenticated GET https://api.tokenfactory.nebius.com/v1/models — body {\"detail\":\"Couldn't authenticate. Reason: token is not present\"}; credential-free, no key used"
+        },
+        "billingProbe": {
+            "date": "2026-09-29",
+            "method": "public page https://docs.tokenfactory.nebius.com/other-capabilities/billing-new — \"Setting up a billing account requires a bank card.\" and \"Upon first sign-up, you receive $1 in trial credit, valid for 30 days.\"; credential-free"
+        },
+        "liveMatrix": null,
+        "addedInPR": "https://github.com/juspay/neurolink/pull/1860"
+    }
+}

package/dist/providers/catalog/prime-intellect.json

@@ -0,0 +1,217 @@
+{
+    "$schema": "./provider-catalog.schema.json",
+    "id": "prime-intellect",
+    "displayName": "Prime Intellect",
+    "aliases": [],
+    "tier": 2,
+    "wire": {
+        "baseURL": "https://api.pinference.ai/api/v1",
+        "apiKeyFallbackEnvVars": ["PRIME_API_KEY"]
+    },
+    "models": {
+        "default": "openai/gpt-4.1-mini",
+        "fallbacks": [
+            "openai/gpt-4.1",
+            "anthropic/claude-sonnet-4.5",
+            "meta-llama/llama-3.3-70b-instruct"
+        ],
+        "defaultContextWindow": 131072,
+        "defaultMaxOutputTokens": 16384,
+        "catalog": {
+            "openai/gpt-4.1-mini": {
+                "contextWindow": 1047576,
+                "maxOutputTokens": 32768,
+                "pricingPerMTok": {
+                    "input": 0.4,
+                    "output": 1.6
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Recommended default. Prime Intellect's own Evaluating Environments guide (docs.primeintellect.ai/tutorials-environments/evaluating) lists this model as \"Fast, cost-effective (default)\" and uses it as the default model of its inference-backed evaluation command. No Prime-trained model appears on this roster (119 of 125 entries are gateway-routed and the other 6 are third-party models served on Prime infrastructure), so this is the model its docs designate as the default, not a Prime-trained flagship. Figures from the unauthenticated GET https://api.pinference.ai/api/v1/models roster (2026-09-29): context_window 1,047,576, max_output_tokens 32,768, $0.4/$1.6 per MTok input/output; input modalities image/text/file; supported_parameters include response_format, structured_outputs, tool_choice, tools; serving gateway. vision: true follows the roster's image input modality (no image request was sent)."
+            },
+            "openai/gpt-4.1": {
+                "contextWindow": 1047576,
+                "maxOutputTokens": 32768,
+                "pricingPerMTok": {
+                    "input": 2,
+                    "output": 8
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Listed in the Evaluating Environments guide (docs.primeintellect.ai/tutorials-environments/evaluating) as \"Higher quality\" than the default. Fallback. Figures from the unauthenticated GET https://api.pinference.ai/api/v1/models roster (2026-09-29): context_window 1,047,576, max_output_tokens 32,768, $2/$8 per MTok input/output; input modalities image/text/file; supported_parameters include response_format, structured_outputs, tool_choice, tools; serving gateway. vision: true follows the roster's image input modality (no image request was sent)."
+            },
+            "anthropic/claude-sonnet-4.5": {
+                "contextWindow": 1000000,
+                "maxOutputTokens": 64000,
+                "pricingPerMTok": {
+                    "input": 3,
+                    "output": 15
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Listed in the Evaluating Environments guide (docs.primeintellect.ai/tutorials-environments/evaluating) as \"Strong reasoning\". Fallback. Figures from the unauthenticated GET https://api.pinference.ai/api/v1/models roster (2026-09-29): context_window 1,000,000, max_output_tokens 64,000, $3/$15 per MTok input/output; input modalities text/image/file; supported_parameters include response_format, structured_outputs, tool_choice, tools; serving gateway. vision: true follows the roster's image input modality (no image request was sent)."
+            },
+            "meta-llama/llama-3.3-70b-instruct": {
+                "contextWindow": 131072,
+                "maxOutputTokens": 16384,
+                "pricingPerMTok": {
+                    "input": 1.04,
+                    "output": 2.253
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Listed in the Evaluating Environments guide (docs.primeintellect.ai/tutorials-environments/evaluating) as \"Open-weight, balanced\". Fallback. Figures from the unauthenticated GET https://api.pinference.ai/api/v1/models roster (2026-09-29): context_window 131,072, max_output_tokens 16,384, $1.04/$2.253 per MTok input/output; input modalities text; supported_parameters include response_format, structured_outputs, tool_choice, tools; serving gateway. No image input modality on the roster, so vision is false."
+            },
+            "google/gemini-2.5-flash": {
+                "contextWindow": 1048576,
+                "maxOutputTokens": 65535,
+                "pricingPerMTok": {
+                    "input": 0.3,
+                    "output": 2.5
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Listed in the Evaluating Environments guide (docs.primeintellect.ai/tutorials-environments/evaluating) as \"Fast multimodal\". Figures from the unauthenticated GET https://api.pinference.ai/api/v1/models roster (2026-09-29): context_window 1,048,576, max_output_tokens 65,535, $0.3/$2.5 per MTok input/output; input modalities file/image/text/audio/video; supported_parameters include response_format, structured_outputs, tool_choice, tools; serving gateway. vision: true follows the roster's image input modality (no image request was sent)."
+            },
+            "Qwen/Qwen3-235B-A22B-Instruct-2507": {
+                "contextWindow": 262144,
+                "maxOutputTokens": 235929,
+                "pricingPerMTok": {
+                    "input": 0.25,
+                    "output": 1
+                },
+                "vision": false,
+                "status": "production",
+                "description": "The Evaluating Environments guide (docs.primeintellect.ai/tutorials-environments/evaluating) lists a lowercase id, qwen/qwen3-235b-a22b-instruct-2507, as \"Large MoE model\"; that lowercase id is not on today's roster, so the id here is copied character for character from the roster's differently cased entry. The roster's max_output_tokens figure is reported as served. Figures from the unauthenticated GET https://api.pinference.ai/api/v1/models roster (2026-09-29): context_window 262,144, max_output_tokens 235,929, $0.25/$1 per MTok input/output; input modalities text; supported_parameters include response_format, structured_outputs, tool_choice, tools; serving gateway. No image input modality on the roster, so vision is false."
+            },
+            "z-ai/glm-5.3": {
+                "contextWindow": 1048576,
+                "maxOutputTokens": 131072,
+                "pricingPerMTok": {
+                    "input": 1.4,
+                    "output": 4.4,
+                    "cachedInput": 0.26
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Serving \"hosted\" on today's roster — the vendor's OpenAPI spec (https://api.pinference.ai/openapi.json) defines that value as models served on Prime infrastructure. Roster display_name \"GLM 5.3\"; reasoning is marked mandatory with default effort \"max\". Its supported_parameters list tools and tool_choice but not response_format or structured_outputs. Figures from the unauthenticated GET https://api.pinference.ai/api/v1/models roster (2026-09-29): context_window 1,048,576, max_output_tokens 131,072, $1.4/$4.4 per MTok input/output ($0.26 cached-input read); input modalities text; supported_parameters include tools, tool_choice; serving hosted. No image input modality on the roster, so vision is false."
+            },
+            "openai/gpt-5.4-mini": {
+                "contextWindow": 400000,
+                "maxOutputTokens": 128000,
+                "pricingPerMTok": {
+                    "input": 0.75,
+                    "output": 4.5
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Named in the Troubleshooting page's direct-API examples (docs.primeintellect.ai/inference/troubleshooting). Roster reasoning efforts: xhigh/high/medium/low/none, not mandatory. Figures from the unauthenticated GET https://api.pinference.ai/api/v1/models roster (2026-09-29): context_window 400,000, max_output_tokens 128,000, $0.75/$4.5 per MTok input/output; input modalities file/image/text; supported_parameters include response_format, structured_outputs, tool_choice, tools; serving gateway. vision: true follows the roster's image input modality (no image request was sent)."
+            },
+            "openai/gpt-oss-120b": {
+                "contextWindow": 131072,
+                "maxOutputTokens": 65536,
+                "pricingPerMTok": {
+                    "input": 0.35,
+                    "output": 0.75
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Roster-listed only. Reasoning is marked mandatory (default effort \"medium\"). Figures from the unauthenticated GET https://api.pinference.ai/api/v1/models roster (2026-09-29): context_window 131,072, max_output_tokens 65,536, $0.35/$0.75 per MTok input/output; input modalities text; supported_parameters include response_format, structured_outputs, tool_choice, tools; serving gateway. No image input modality on the roster, so vision is false."
+            },
+            "deepseek/deepseek-v4-flash": {
+                "contextWindow": 1048576,
+                "maxOutputTokens": 131072,
+                "pricingPerMTok": {
+                    "input": 0.44,
+                    "output": 1.32
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Roster-listed only. Figures from the unauthenticated GET https://api.pinference.ai/api/v1/models roster (2026-09-29): context_window 1,048,576, max_output_tokens 131,072, $0.44/$1.32 per MTok input/output; input modalities text; supported_parameters include response_format, structured_outputs, tool_choice, tools; serving gateway. No image input modality on the roster, so vision is false."
+            },
+            "anthropic/claude-haiku-4.5": {
+                "contextWindow": 200000,
+                "maxOutputTokens": 64000,
+                "pricingPerMTok": {
+                    "input": 1,
+                    "output": 5
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Roster-listed only. Figures from the unauthenticated GET https://api.pinference.ai/api/v1/models roster (2026-09-29): context_window 200,000, max_output_tokens 64,000, $1/$5 per MTok input/output; input modalities text/image/file; supported_parameters include response_format, structured_outputs, tool_choice, tools; serving gateway. vision: true follows the roster's image input modality (no image request was sent)."
+            },
+            "Qwen/Qwen3.5-122B-A10B": {
+                "pricingPerMTok": {
+                    "input": 0.3,
+                    "output": 0.9,
+                    "cachedInput": 0.27
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Serving \"hosted\" (Prime infrastructure) on today's roster. The roster gives it no context_window, max_output_tokens, modalities or supported_parameters, so none are declared here; vision false is the schema's required default, not a vendor statement about image input. Roster-listed only. Figures from the unauthenticated GET https://api.pinference.ai/api/v1/models roster (2026-09-29): $0.3/$0.9 per MTok input/output ($0.27 cached-input read); serving hosted."
+            }
+        },
+        "topModels": [
+            "openai/gpt-4.1-mini",
+            "openai/gpt-4.1",
+            "anthropic/claude-sonnet-4.5",
+            "meta-llama/llama-3.3-70b-instruct",
+            "google/gemini-2.5-flash",
+            "Qwen/Qwen3-235B-A22B-Instruct-2507",
+            "z-ai/glm-5.3",
+            "openai/gpt-5.4-mini",
+            "openai/gpt-oss-120b",
+            "deepseek/deepseek-v4-flash",
+            "anthropic/claude-haiku-4.5",
+            "Qwen/Qwen3.5-122B-A10B"
+        ]
+    },
+    "capabilities": {
+        "text": true,
+        "streaming": true,
+        "tools": "model-dependent",
+        "toolsWithStreaming": false,
+        "structuredOutput": true,
+        "structuredOutputWithTools": false,
+        "embeddings": false,
+        "thinking": false
+    },
+    "errorRules": [
+        {
+            "status": 401,
+            "pattern": "Authorization header or x-api-key missing|invalid_api_key|Invalid API key",
+            "class": "authentication",
+            "message": "Invalid or missing Prime Intellect API key. Check {apiKeyEnvVar}. Create a key with the Inference permission enabled at {setupUrl}"
+        }
+    ],
+    "setup": {
+        "url": "https://app.primeintellect.ai/dashboard/tokens",
+        "apiKeyFormat": null,
+        "billingPolicy": "no-free-tier",
+        "instructions": [
+            "1. Visit: https://app.primeintellect.ai/dashboard/tokens (Settings -> API Keys) and sign in",
+            "2. Click Generate New Key and enable the Inference permission — without it requests fail with authentication errors (https://docs.primeintellect.ai/inference/overview); the key is shown only once, so copy it straight away (https://docs.primeintellect.ai/api-reference/api-keys)",
+            "3. Billing is automatic deduction from your Prime Intellect account balance (https://docs.primeintellect.ai/inference/overview); no free tier or trial credit is documented on the pages read, so add funds at https://app.primeintellect.ai/dashboard/billing and confirm current terms before the first call",
+            "4. Set {apiKeyEnvVar} in your .env file (PRIME_API_KEY, the variable Prime's own docs use, is also read)"
+        ],
+        "description": "OpenAI-compatible endpoint at https://api.pinference.ai/api/v1 (docs.primeintellect.ai/inference/overview), a gateway that routes to upstream model providers plus a few models served on Prime infrastructure. The inference host (pinference.ai) differs from the docs and dashboard host (primeintellect.ai); both are confirmed on the vendor's own pages. The catalog schema has no 'unknown' billing value, so no-free-tier here means no free tier is documented, not that the vendor confirmed there is none."
+    },
+    "evidence": {
+        "rosterVerified": {
+            "date": "2026-09-29",
+            "method": "unauthenticated GET https://api.pinference.ai/api/v1/models — 125 models; credential-free, no key used",
+            "status": 200
+        },
+        "authProbe": {
+            "date": "2026-09-29",
+            "method": "unauthenticated GET https://api.pinference.ai/api/v1/models/openai/gpt-4.1-mini with no credentials — 401 {\"detail\":\"Authorization header or x-api-key missing\", ...}; only the missing-credential shape was observed, the invalid-key shape (error.code invalid_api_key) is from docs.primeintellect.ai/api-reference/inference-models and was not probed; credential-free, no key used",
+            "status": 401
+        },
+        "billingProbe": {
+            "date": "2026-09-29",
+            "method": "public page https://docs.primeintellect.ai/inference/overview — \"Automatic deduction from your Prime Intellect account balance\"; https://docs.primeintellect.ai/inference/troubleshooting documents an insufficient_funds error; no free tier stated; credential-free"
+        },
+        "liveMatrix": null,
+        "addedInPR": "https://github.com/juspay/neurolink/pull/1860"
+    }
+}