@juspay/neurolink 12.38.1 → 12.39.1

package/dist/neurolink.js

@@ -4250,6 +4250,12 @@ Current user's request: ${currentInput}`;
             provider: options.provider,
             model: options.model,
             temperature: options.temperature,
+            // Listed explicitly because this literal is the only road into the
+            // provider: `topP` is declared on both `GenerateOptions` and
+            // `TextGenerationOptions`, but it was missing here and so was silently
+            // dropped on every generate() call regardless of provider, while
+            // stream() (which spreads its options) already forwarded it.
+            topP: options.topP,
             maxTokens: options.maxTokens,
             systemPrompt: options.systemPrompt,
             schema: options.schema,

package/dist/providers/catalog/above-dev.json

@@ -0,0 +1,194 @@
+{
+    "$schema": "./provider-catalog.schema.json",
+    "id": "above-dev",
+    "displayName": "above.dev",
+    "aliases": [],
+    "tier": 2,
+    "wire": {
+        "baseURL": "https://api.above.dev/v1"
+    },
+    "models": {
+        "default": "deepseek-v4.1-flash",
+        "fallbacks": [
+            "glm-5.3-flash",
+            "mimo-v2.6-flash",
+            "deepseek-v4-pro",
+            "glm-5.2",
+            "mimo-v2.6-pro",
+            "glm-5.2-fast",
+            "qwen3.8-max",
+            "mimo-v2.6-pro-ultraspeed"
+        ],
+        "defaultContextWindow": 1000000,
+        "defaultMaxOutputTokens": 131072,
+        "catalog": {
+            "deepseek-v4.1-flash": {
+                "contextWindow": 1000000,
+                "maxOutputTokens": 131072,
+                "vision": true,
+                "status": "production",
+                "description": "DeepSeek V4.1 Flash (also addressable as deepseek-v4-flash), tier T1. above.dev's own default: naming no model routes here (\"the cheapest model in the catalog\", https://above.dev/docs, retrieved 2026-09-28). GET /v1/models (unauthenticated, 2026-09-28) self-reports supports_tool_calls, supports_streaming, supports_json_mode, supports_structured_output and supports_vision all true, and supports_reasoning true, but none of this was independently probed — no key was used. Pricing uses above.dev's documented time-of-day split (peak 01:00-04:00 and 06:00-10:00 UTC weekdays, else off-peak; weekends fully off-peak in Beijing time) so pricingPerMTok is omitted rather than collapsed to one number: input $0.165-$0.33/M, cache-hit $0.0033-$0.0066/M, output $0.66-$1.32/M (https://above.dev/docs, retrieved 2026-09-28)."
+            },
+            "deepseek-v4-pro": {
+                "contextWindow": 1000000,
+                "maxOutputTokens": 131072,
+                "vision": false,
+                "status": "production",
+                "description": "DeepSeek V4 Pro, tier T2 (\"Agentic — multi-step, tool-capable\" per the docs' tier table). Roster-listed via unauthenticated GET /v1/models (2026-09-28); not capability-probed. Time-of-day pricing per https://above.dev/docs (retrieved 2026-09-28) — input $0.726-$1.452/M, cache-hit $0.0242-$0.0484/M, output $2.178-$4.356/M — so pricingPerMTok is omitted (see deepseek-v4.1-flash for why)."
+            },
+            "glm-5.2": {
+                "contextWindow": 1048576,
+                "maxOutputTokens": 131072,
+                "pricingPerMTok": {
+                    "input": 1.54,
+                    "output": 4.84,
+                    "cachedInput": 0.154
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Zhipu GLM 5.2, tier T2. Roster-listed via unauthenticated GET /v1/models (2026-09-28); not capability-probed. Flat (non-time-of-day) pricing per https://above.dev/docs, retrieved 2026-09-28."
+            },
+            "glm-5.3-flash": {
+                "contextWindow": 1048576,
+                "maxOutputTokens": 131072,
+                "pricingPerMTok": {
+                    "input": 0.165,
+                    "output": 0.55,
+                    "cachedInput": 0.0319
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Zhipu GLM 5.3 Flash, tier T1. Roster-listed via unauthenticated GET /v1/models (2026-09-28), which self-reports supports_vision true for this id; not capability-probed with a real image. Flat pricing per https://above.dev/docs, retrieved 2026-09-28. Note: a separate id, glm-5.3-flash-modal, is named in the docs' Limits section as one of two \"free-tier routes\" capped at 16,384 output tokens — that id does not appear in the current /v1/models roster and is a distinct route from this one; not represented in this catalog."
+            },
+            "glm-5.2-fast": {
+                "contextWindow": 1048576,
+                "maxOutputTokens": 131072,
+                "pricingPerMTok": {
+                    "input": 2.31,
+                    "output": 7.26,
+                    "cachedInput": 0.231
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Zhipu GLM 5.2 Fast, tier T3 (\"Highest capability in the catalog\" per the docs' tier table). Roster-listed via unauthenticated GET /v1/models (2026-09-28); not capability-probed. Flat pricing per https://above.dev/docs, retrieved 2026-09-28."
+            },
+            "qwen3.8-max": {
+                "contextWindow": 1000000,
+                "maxOutputTokens": 131072,
+                "pricingPerMTok": {
+                    "input": 2.2,
+                    "output": 6.6,
+                    "cachedInput": 0.275
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Alibaba Qwen 3.8 Max, tier T3. Roster-listed via unauthenticated GET /v1/models (2026-09-28), which self-reports supports_vision true for this id; not capability-probed with a real image. Flat pricing per https://above.dev/docs, retrieved 2026-09-28."
+            },
+            "mimo-v2.6-pro": {
+                "contextWindow": 1000000,
+                "maxOutputTokens": 131072,
+                "pricingPerMTok": {
+                    "input": 0.5077,
+                    "output": 1.0154,
+                    "cachedInput": 0.0042
+                },
+                "vision": false,
+                "status": "production",
+                "description": "MiMo V2.6 Pro (also addressable as mimo-v2.5-pro), tier T2. Roster-listed via unauthenticated GET /v1/models (2026-09-28); not capability-probed. Flat pricing per https://above.dev/docs, retrieved 2026-09-28."
+            },
+            "mimo-v2.6-flash": {
+                "contextWindow": 1000000,
+                "maxOutputTokens": 131072,
+                "pricingPerMTok": {
+                    "input": 0.1692,
+                    "output": 0.3385,
+                    "cachedInput": 0.0034
+                },
+                "vision": false,
+                "status": "production",
+                "description": "MiMo V2.6 Flash, tier T1. Roster-listed via unauthenticated GET /v1/models (2026-09-28); not capability-probed. Flat pricing per https://above.dev/docs, retrieved 2026-09-28."
+            },
+            "mimo-v2.6-pro-ultraspeed": {
+                "contextWindow": 1000000,
+                "maxOutputTokens": 131072,
+                "pricingPerMTok": {
+                    "input": 5.0769,
+                    "output": 10.1538,
+                    "cachedInput": 0.0423
+                },
+                "vision": false,
+                "status": "production",
+                "description": "MiMo V2.6 Pro UltraSpeed, tier T3. Roster-listed via unauthenticated GET /v1/models (2026-09-28); not capability-probed. Flat pricing per https://above.dev/docs, retrieved 2026-09-28."
+            }
+        },
+        "topModels": [
+            "deepseek-v4.1-flash",
+            "glm-5.3-flash",
+            "mimo-v2.6-flash",
+            "deepseek-v4-pro",
+            "glm-5.2",
+            "mimo-v2.6-pro",
+            "glm-5.2-fast",
+            "qwen3.8-max",
+            "mimo-v2.6-pro-ultraspeed"
+        ]
+    },
+    "capabilities": {
+        "text": true,
+        "streaming": true,
+        "tools": "model-dependent",
+        "toolsWithStreaming": false,
+        "structuredOutput": false,
+        "structuredOutputWithTools": false,
+        "embeddings": false,
+        "thinking": false
+    },
+    "errorRules": [
+        {
+            "status": 401,
+            "class": "authentication",
+            "message": "Invalid or missing above.dev API key (401 Unauthorised). Check {apiKeyEnvVar}. Get one at {setupUrl}"
+        },
+        {
+            "status": 402,
+            "class": "provider",
+            "message": "above.dev account balance is insufficient (402 payment required). Top up at https://above.dev/pricing"
+        },
+        {
+            "status": 403,
+            "class": "invalid-model",
+            "message": "above.dev model '{model}' is not enabled for this key (403 model_not_allowed_for_key) — it may be plan-restricted or in private beta. Pick another model or contact above.dev."
+        },
+        {
+            "status": 429,
+            "class": "rate-limit",
+            "message": "above.dev rate limit exceeded (429 rate_limit_exceeded, 300 requests/minute per account). Retry after the documented Retry-After header."
+        },
+        {
+            "status": 504,
+            "class": "network",
+            "message": "above.dev upstream model did not respond in time (504 provider_timeout). Retry the request."
+        }
+    ],
+    "setup": {
+        "url": "https://above.dev/pricing",
+        "apiKeyFormat": "^sk-gw-.+$",
+        "billingPolicy": "no-free-tier",
+        "instructions": [
+            "1. Visit: https://above.dev/pricing and purchase credit — API keys are issued only after purchase, there is no free signup",
+            "2. Copy the issued key — it starts with sk-gw- (the docs give only the prefix, not a full character-set spec)",
+            "3. Manage or rotate the key from https://above.dev/dashboard if it is ever compromised — a key grants full access to your credit balance",
+            "4. Set {apiKeyEnvVar} in your .env file"
+        ],
+        "description": "OpenAI-compatible gateway at https://api.above.dev/v1, routing to a small DeepSeek/GLM/Qwen/MiMo catalog. POST /v1/chat/completions is documented as an alias that delegates to the vendor's native POST /v1/messages with \"identical behaviour\". No free tier for account access — a key is issued only after purchase at above.dev/pricing (the docs separately mention two lower-cap \"free-tier routes\", glm-5.3-flash-modal and deepseek-v4.1-flash-modal, but those ids are outside the current /v1/models roster and still require a purchased key). Account-wide rate limit is 300 requests/minute; every current model accepts up to 131,072 output tokens. The docs describe the `tools` and `response_format` request fields only as routing signals (\"presence influences routing toward tool-capable models\" / \"scores toward structured-output models\"), in explicit contrast to `tool_choice` and `temperature`, which the same table states are \"passed through to the model\" — so this entry does not assert tool calling or structured output as a confirmed, working capability without a live probe."
+    },
+    "evidence": {
+        "rosterVerified": {
+            "date": "2026-09-28",
+            "method": "unauthenticated GET https://api.above.dev/v1/models — 9 models; credential-free, no key used",
+            "status": 200
+        },
+        "liveMatrix": null,
+        "addedInPR": "https://github.com/juspay/neurolink/pull/1848"
+    }
+}

package/dist/providers/catalog/aiand.json

@@ -0,0 +1,204 @@
+{
+    "$schema": "./provider-catalog.schema.json",
+    "id": "aiand",
+    "displayName": "ai&",
+    "aliases": [],
+    "tier": 2,
+    "wire": {
+        "baseURL": "https://api.aiand.com/v1"
+    },
+    "models": {
+        "default": "openai/gpt-oss-120b",
+        "fallbacks": [
+            "zai-org/glm-5.3",
+            "zai-org/glm-5.2",
+            "deepseek-ai/deepseek-v4-pro",
+            "deepseek-ai/deepseek-v4-flash",
+            "moonshotai/kimi-k3",
+            "moonshotai/kimi-k2.7-code",
+            "qwen/qwen3.8-27b",
+            "qwen/qwen3.6-27b",
+            "motif-technologies/motif-3",
+            "google/gemma-4-31b-it"
+        ],
+        "defaultContextWindow": 131072,
+        "defaultMaxOutputTokens": 131072,
+        "catalog": {
+            "openai/gpt-oss-120b": {
+                "contextWindow": 131072,
+                "pricingPerMTok": {
+                    "input": 0.15,
+                    "output": 0.6,
+                    "cachedInput": 0.08
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Default — the model used in ai&'s own quickstart curl example (https://docs.aiand.com/, retrieved 2026-09-28): 'curl https://api.aiand.com/v1/chat/completions -H \"Authorization: Bearer sk-your-api-key\" -d {\"model\": \"openai/gpt-oss-120b\", ...}'. Vendor description (GET /v1/models, 2026-09-28): \"Open GPT reasoning model for self-hosted agents and controllable deployments\". Cheapest model on the roster. Documents `reasoning` + `tool_calling` capabilities and a `reasoning_effort` request parameter with model-specific values low/medium/high (default medium) — see https://docs.aiand.com/api/chat-completions/. No `vision` capability listed, so image/video input is expected to be rejected. Docs- and roster-verified only — no API key was used, so no live chat/stream/tool call was made against this or any other model in this catalog."
+            },
+            "zai-org/glm-5.3": {
+                "contextWindow": 1048576,
+                "pricingPerMTok": {
+                    "input": 1.0,
+                    "output": 4.0,
+                    "cachedInput": 0.3
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Vendor description (GET /v1/models, 2026-09-28): \"Flagship GLM model for long-horizon coding, agents, and complex project delivery\". reasoning_effort values low/high/max, default max. 1M context. Roster-listed and priced only, not itself capability-probed; fallback."
+            },
+            "zai-org/glm-5.2": {
+                "contextWindow": 1048576,
+                "pricingPerMTok": {
+                    "input": 1.0,
+                    "output": 4.0,
+                    "cachedInput": 0.3
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Vendor description (GET /v1/models, 2026-09-28): \"Open flagship GLM for long-horizon coding agents and million-token context work\". reasoning_effort values none/high/max, default max. Roster-listed only; fallback."
+            },
+            "deepseek-ai/deepseek-v4-pro": {
+                "contextWindow": 1048576,
+                "pricingPerMTok": {
+                    "input": 1.0,
+                    "output": 2.5,
+                    "cachedInput": 0.25
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Vendor description (GET /v1/models, 2026-09-28): \"Open MoE flagship with million-token context for coding and long agent runs\". reasoning_effort values none/high/max, default none. Roster-listed only; fallback."
+            },
+            "deepseek-ai/deepseek-v4-flash": {
+                "contextWindow": 1048576,
+                "pricingPerMTok": {
+                    "input": 0.15,
+                    "output": 0.25,
+                    "cachedInput": 0.08
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Vendor description (GET /v1/models, 2026-09-28): \"Fast DeepSeek V4 lane for economical reasoning, coding, and long-context work\". reasoning_effort values none/high/max, default none. Roster-listed only; fallback."
+            },
+            "moonshotai/kimi-k3": {
+                "contextWindow": 1048576,
+                "pricingPerMTok": {
+                    "input": 3.0,
+                    "output": 12.5,
+                    "cachedInput": 0.5
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Vendor description (GET /v1/models, 2026-09-28): \"Multimodal Kimi model with 1M context and toggleable max-effort thinking for long-horizon agent work\". reasoning_effort values low/high/max, default max. Lists `vision` in its capabilities array, but ai&'s vision docs (https://docs.aiand.com/capabilities/vision/) describe images/video as uploaded via the Files API first and referenced by `file_id` — not the inline base64 `image_url` data-URI shape NeuroLink's other providers use — and that flow was not exercised (no key used). Roster-listed only; fallback."
+            },
+            "moonshotai/kimi-k2.7-code": {
+                "contextWindow": 262144,
+                "pricingPerMTok": {
+                    "input": 0.75,
+                    "output": 3.5,
+                    "cachedInput": 0.2
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Vendor description (GET /v1/models, 2026-09-28): \"Coding-focused Kimi model, stronger on long-horizon repo work with less overthinking\". reasoning_effort fixed to high (only value listed). Lists `vision`; see the kimi-k3 note above on ai&'s Files-API-based image/video flow being unverified here. Roster-listed only; fallback."
+            },
+            "qwen/qwen3.8-27b": {
+                "contextWindow": 262144,
+                "pricingPerMTok": {
+                    "input": 0.4,
+                    "output": 3.0,
+                    "cachedInput": 0.2
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Vendor description (GET /v1/models, 2026-09-28): \"Dense 27B vision-language model for coding, agent tasks, and image and video understanding\". reasoning_effort values none/low/medium/xhigh, default medium. Lists `vision`; see the kimi-k3 note above on ai&'s Files-API-based image/video flow being unverified here. Roster-listed only; fallback."
+            },
+            "qwen/qwen3.6-27b": {
+                "contextWindow": 262144,
+                "pricingPerMTok": {
+                    "input": 0.32,
+                    "output": 3.2,
+                    "cachedInput": 0.2
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Vendor description (GET /v1/models, 2026-09-28): \"Qwen vision-language model for visual reasoning, documents, and agent tasks\". reasoning_effort values none/high, default high. Lists `vision`; see the kimi-k3 note above on ai&'s Files-API-based image/video flow being unverified here. Roster-listed only; fallback."
+            },
+            "motif-technologies/motif-3": {
+                "contextWindow": 262144,
+                "pricingPerMTok": {
+                    "input": 0.5,
+                    "output": 2.0,
+                    "cachedInput": 0.2
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Vendor description (GET /v1/models, 2026-09-28): \"Motif 3 is a large-scale, decoder-only Mixture-of-Experts (MoE) language model with 314 billion total parameters and 13.2 billion parameters activated per token.\" reasoning_effort values none/high, default high. Roster-listed only; fallback."
+            },
+            "google/gemma-4-31b-it": {
+                "contextWindow": 262144,
+                "pricingPerMTok": {
+                    "input": 0.2,
+                    "output": 0.5,
+                    "cachedInput": 0.05
+                },
+                "vision": true,
+                "status": "production",
+                "description": "Vendor description (GET /v1/models, 2026-09-28): \"Largest Gemma 4 instruction model for open, self-hosted chat and reasoning\". reasoning_effort values none/high, default none. Lists `vision`; see the kimi-k3 note above on ai&'s Files-API-based image/video flow being unverified here. Roster-listed only; fallback."
+            }
+        },
+        "topModels": [
+            "openai/gpt-oss-120b",
+            "zai-org/glm-5.3",
+            "deepseek-ai/deepseek-v4-flash",
+            "moonshotai/kimi-k3",
+            "qwen/qwen3.8-27b",
+            "zai-org/glm-5.2",
+            "google/gemma-4-31b-it"
+        ]
+    },
+    "capabilities": {
+        "text": true,
+        "streaming": true,
+        "tools": true,
+        "toolsWithStreaming": true,
+        "structuredOutput": true,
+        "structuredOutputWithTools": false,
+        "embeddings": false,
+        "thinking": true
+    },
+    "errorRules": [
+        {
+            "status": 401,
+            "pattern": "invalid_api_key|authentication_error",
+            "class": "authentication",
+            "message": "Invalid ai& API key. Check {apiKeyEnvVar}. Get one at {setupUrl}"
+        },
+        {
+            "status": 429,
+            "pattern": "rate_limit_exceeded|rate_limit_error",
+            "class": "rate-limit",
+            "message": "ai& rate limit exceeded. Check the X-RateLimit-Policy response header and retry with backoff, or review your plan at {setupUrl}"
+        }
+    ],
+    "setup": {
+        "url": "https://console.aiand.com/playground",
+        "apiKeyFormat": null,
+        "billingPolicy": "no-free-tier",
+        "instructions": [
+            "1. Visit: https://console.aiand.com/playground and sign in (Google, GitHub, or a magic-link email — https://docs.aiand.com/authentication/)",
+            "2. Create an organization API key under Organizations → API Keys (https://docs.aiand.com/organizations/api-keys/) — shown in full only once, and starts with sk-",
+            "3. ai& is prepaid only, no free tier: add credits via Stripe Checkout before making requests (https://docs.aiand.com/billing/credits/) — a zero balance returns HTTP 402 until credits are added",
+            "4. Set {apiKeyEnvVar} in your .env file"
+        ],
+        "description": "OpenAI-compatible endpoint at https://api.aiand.com/v1 (also exposes Anthropic-shaped /v1/messages, a Responses endpoint, and legacy Completions), serving zai-org GLM, Qwen, Kimi (moonshotai), DeepSeek V4, Google Gemma 4, Motif-3 and OpenAI gpt-oss open-weight models, each with a documented per-model `reasoning_effort` control. Prepaid credit billing only — no free tier; failed requests are not billed."
+    },
+    "evidence": {
+        "rosterVerified": {
+            "date": "2026-09-28",
+            "method": "unauthenticated GET https://api.aiand.com/v1/models — 11 models; credential-free, no key used",
+            "status": 200
+        },
+        "liveMatrix": null,
+        "addedInPR": "https://github.com/juspay/neurolink/pull/1848"
+    }
+}

package/dist/providers/catalog/ambient.json

@@ -0,0 +1,117 @@
+{
+    "$schema": "./provider-catalog.schema.json",
+    "id": "ambient",
+    "displayName": "Ambient",
+    "aliases": [],
+    "tier": 2,
+    "wire": {
+        "baseURL": "https://api.ambient.xyz/v1"
+    },
+    "models": {
+        "default": "ambient/large",
+        "fallbacks": ["z-ai/glm-5.2", "qwen/qwen3.6-27b", "qwen/qwen3.8-27b"],
+        "defaultContextWindow": 202752,
+        "defaultMaxOutputTokens": 202752,
+        "catalog": {
+            "ambient/large": {
+                "contextWindow": 202752,
+                "maxOutputTokens": 202752,
+                "pricingPerMTok": {
+                    "input": 0.6,
+                    "output": 2.0,
+                    "cachedInput": 0.15
+                },
+                "vision": false,
+                "status": "production",
+                "description": "Ambient's own recommended default — the model id used in every example on the vendor's docs (the plain chat-completions curl example and the streaming curl example both send model: \"ambient/large\"). Documented as \"an Ambient-managed alias rather than a distinct model. Its specs mirror whatever concrete model it currently points to\" (https://ambient.xyz/docs/build/models, retrieved 2026-09-28); today it mirrors z-ai/glm-5.2 exactly — identical context_length, max_output_length and pricing in the unauthenticated GET https://api.ambient.xyz/v1/models response taken 2026-09-28. Text-only (input_modalities: [\"text\"]); no vision. The live roster's supported_features for this id are tools, json_mode, structured_outputs, reasoning, logprobs — this is the vendor's own server-reported metadata, not a live POST probe (none was possible credential-free), so capabilities.tools and capabilities.structuredOutput are marked true on that basis alone, while the combined capabilities (tools+response_format in one request, tool-call deltas under streaming) are unverified and both structuredOutputWithTools and toolsWithStreaming are left false pending a live-keyed probe. maxOutputTokens equals the context window in the vendor's own catalog entry (202,752 tokens each, i.e. \"198K\" as rounded on the docs models page)."
+            },
+            "z-ai/glm-5.2": {
+                "contextWindow": 202752,
+                "maxOutputTokens": 202752,
+                "pricingPerMTok": {
+                    "input": 0.6,
+                    "output": 2.0,
+                    "cachedInput": 0.15
+                },
+                "vision": false,
+                "status": "production",
+                "description": "The concrete underlying model that ambient/large currently aliases. The vendor's docs models page states \"GLM 5.2 ... is the supported production model\" (https://ambient.xyz/docs/build/models, retrieved 2026-09-28); other catalog ids may be previews with no guarantees. Text-only, no vision. Same live-roster numbers (unauthenticated GET https://api.ambient.xyz/v1/models, 2026-09-28) and the same unverified-combined-capability caveats as ambient/large — kept as a fallback pinned to a concrete id in case the ambient/large alias target changes underneath it."
+            },
+            "qwen/qwen3.6-27b": {
+                "contextWindow": 32768,
+                "maxOutputTokens": 8192,
+                "pricingPerMTok": {
+                    "input": 0.32,
+                    "output": 3.2,
+                    "cachedInput": 0.16
+                },
+                "vision": true,
+                "status": "preview",
+                "description": "Qwen3.6 27B. The live roster's input_modalities are [\"text\",\"image\"] and the vendor's own description reads \"natively multimodal (text + image) dense 27B, hybrid attention. Served FP8 at 32k context, thinking off.\" (unauthenticated GET https://api.ambient.xyz/v1/models, 2026-09-28). The vendor's docs models page marks non-GLM-5.2 entries as possible previews served with no guarantees, so status is preview here. vision: true is taken from input_modalities; set as models.visionModel since the default (ambient/large) is text-only. Same unverified-combined-capability caveats as the other entries (no live POST probe was possible)."
+            },
+            "qwen/qwen3.8-27b": {
+                "contextWindow": 32768,
+                "maxOutputTokens": 8192,
+                "pricingPerMTok": {
+                    "input": 0.32,
+                    "output": 3.2,
+                    "cachedInput": 0.16
+                },
+                "vision": true,
+                "status": "preview",
+                "description": "Qwen3.8 27B. The live roster's own description calls it \"natively multimodal (text + image + video) dense 27B, Apache 2.0, hybrid attention\" but input_modalities lists only [\"text\",\"image\"] — no \"video\" entry — so vision: true here is based on input_modalities, not the prose \"video\" claim, and no video-input support is declared. Same preview status and unverified-combined-capability caveats as qwen3.6-27b (unauthenticated GET https://api.ambient.xyz/v1/models, 2026-09-28)."
+            }
+        },
+        "topModels": [
+            "ambient/large",
+            "z-ai/glm-5.2",
+            "qwen/qwen3.6-27b",
+            "qwen/qwen3.8-27b"
+        ],
+        "visionModel": "qwen/qwen3.6-27b"
+    },
+    "capabilities": {
+        "text": true,
+        "streaming": true,
+        "tools": true,
+        "toolsWithStreaming": false,
+        "structuredOutput": true,
+        "structuredOutputWithTools": false,
+        "embeddings": false,
+        "thinking": true
+    },
+    "errorRules": [
+        {
+            "status": 401,
+            "class": "authentication",
+            "message": "Invalid Ambient API key. Check {apiKeyEnvVar}. Get one at {setupUrl}"
+        },
+        {
+            "status": 400,
+            "pattern": "Unknown model",
+            "class": "invalid-model",
+            "message": "Ambient model '{model}' not found. Pick a current model from the unauthenticated GET https://api.ambient.xyz/v1/models roster or {setupUrl}"
+        }
+    ],
+    "setup": {
+        "url": "https://app.ambient.xyz",
+        "apiKeyFormat": null,
+        "billingPolicy": "no-free-tier",
+        "instructions": [
+            "1. Visit: https://app.ambient.xyz and sign in",
+            "2. Click \"Get API key\" to create a key",
+            "3. Ambient has no free tier — the cheapest option is the $1/mo Starter plan (https://ambient.xyz/pricing); confirm current billing terms before use",
+            "4. Set {apiKeyEnvVar} in your .env file"
+        ],
+        "description": "OpenAI- and Anthropic-compatible endpoint at https://api.ambient.xyz/v1 (vendor's own docs: \"Point your OpenAI or Anthropic SDK at Ambient and get verified inference in minutes, no rewrite required\"). Pay-per-token API billing; no free tier. Inference is cryptographically verified via the vendor's \"Proof of Logits\" protocol."
+    },
+    "evidence": {
+        "rosterVerified": {
+            "date": "2026-09-28",
+            "method": "unauthenticated GET https://api.ambient.xyz/v1/models — 4 models; credential-free, no key used",
+            "status": 200
+        },
+        "liveMatrix": null,
+        "addedInPR": "https://github.com/juspay/neurolink/pull/1848"
+    }
+}