@kortix/llm-catalog 0.13.30 → 0.13.31

package/README.md

@@ -4,21 +4,67 @@ This package supplies the bundled provider catalog and the Kortix managed model
 
 ## Managed models
 
-The managed lineup uses OpenRouter routes pinned to named zero-data-retention endpoints. The model picker groups every managed model under **Kortix**. Requests use Kortix credits. Project BYOK providers remain separate.
+The picker groups every managed model under **Kortix**, and the provider is always Kortix. Requests use Kortix credits. Project BYOK providers remain separate.
 
-| Picker name | Gateway model ID | Input | USD per 1M input / cached input / output tokens |
-| --- | --- | --- | --- |
-| DeepSeek V4.1 Flash (default) | `deepseek-v4.1-flash` | Text, image | $0.20 / $0.006 / $0.60 |
-| GLM-5.3-Flash | `glm-5.3-flash` | Text, image | $0.15 / $0.05 / $0.50 |
-| Kimi K3 2.8T | `kimi-k3` | Text, image | $2.50 / $0.25 / $10.95 |
+| Picker name | Gateway model ID | Morph model ID | Input | USD per 1M input / cached input / output tokens |
+| --- | --- | --- | --- | --- |
+| DeepSeek V4.1 Flash (default) | `deepseek-v4.1-flash` | `morph-dsv41flash` | Text, image | $0.15 / $0.0359375 / $0.60 |
+| GLM-5.3-Flash | `glm-5.3-flash` | `morph-glm53flash` | Text, image | $0.10 / $0.02 / $0.35 |
+| Kimi K3 2.8T | `kimi-k3` | `morph-kimik3` | Text, image | $2.50 / $0.29 / $14.00 |
 
-The prices match the pinned [OpenRouter endpoints](https://openrouter.ai/api/v1/endpoints/zdr) on 2026-09-18. All rates exclude Kortix credit markup.
+The prices are Morph's list prices on 2026-09-24 (`https://docs.morphllm.com/llms.txt`). Morph publishes no cached rate for DeepSeek V4.1 Flash; the table uses its DeepSeek V4 Flash cached rate. All rates exclude Kortix credit markup.
 
-The OpenCode reference is `kortix/<gateway model ID>`. The bundled sandbox fallback uses the same IDs and capabilities. All entries require `OPENROUTER_API_KEY`.
+The OpenCode reference is `kortix/<gateway model ID>`. The bundled sandbox fallback (`MINIMAL_FALLBACK_MODELS`) uses the same IDs, prices, and capabilities.
 
-DeepSeek V4.1 Flash uses `deepinfra/fp8`; Kimi K3 uses `wafer`; GLM-5.3-Flash uses `coreweave/nvfp4`. Every route sets `zdr: true`, `data_collection: deny`, and `allow_fallbacks: false`. All three exact model/endpoint pairs appeared in OpenRouter's ZDR endpoint feed on 2026-09-18. Kimi K3 returned HTTP 200 for text and image requests on the pinned Wafer route. On 2026-09-18, pinned GLM requests returned HTTP 429 `rate_limit_exceeded` from CoreWeave's shared pool. DeepSeek V4.1 Flash remains the default until a pinned GLM text and image request succeeds with the deployment key.
+### Routing: Morph first, OpenRouter pool second
 
-The public Morph feed lists a different GLM-5.3-Flash endpoint (`morph-glm53flash`) at $0.10 / $0.02 / $0.35. On 2026-09-17, the supplied Morph key received HTTP 400 `invalid_request_error` for that ID. MiniMax M3 and Qwen 3.8 27B are absent from that feed and returned HTTP 400. GLM-5.3 744B accepts only text input and is excluded.
+1. The gateway sends the request to Morph's OpenAI-compatible API (`MORPH_API_URL`, `MORPH_API_KEY`) with the Morph model ID.
+2. When Morph fails before any output (non-2xx status or network error), the gateway sends the same request to OpenRouter (`OPENROUTER_API_KEY`) with the OpenRouter model ID. The descriptor field `failover: true` enables this; `packages/llm-gateway/src/pipeline/simple-handler.ts` implements it. BYOK descriptors never set it.
+3. OpenRouter routes inside the model's endpoint pool: `only` lists the pool, `allow_fallbacks: true` lets OpenRouter move to the next pool member on an endpoint error, `zdr: true` and `data_collection: deny` are forced by the gateway, and `max_price` (USD per 1M tokens) excludes premium tiers.
+
+A failure after output has started is not retried. The client receives the stream error.
+
+Billing: a Morph request bills the table above. An OpenRouter request bills OpenRouter's reported `usage.cost` for the endpoint that served it, so a fallback costs at most `max_price`.
+
+A deployment with only one of the two keys serves managed models through that provider alone.
+
+### OpenRouter fallback pools
+
+Every pool member has a **confirmed US datacenter**. OpenRouter lists the provider's headquarters AND datacenters as US (`/api/v1/providers`), or the endpoint tag names the US region (`/us`). US headquarters alone does not qualify. Every member is also in OpenRouter's ZDR endpoint feed (`/api/v1/endpoints/zdr`), checked on 2026-09-24.
+
+| Model | Pool (`only`) | `max_price` prompt / completion |
+| --- | --- | --- |
+| DeepSeek V4.1 Flash | `morph`, `coreweave/fp8` | $0.30 / $1.20 |
+| GLM-5.3-Flash | `morph`, `decart/fp4`, `coreweave/nvfp4`, `sail-research/us` | $0.15 / $0.50 |
+| Kimi K3 2.8T | `morph`, `fireworks/us` | $3.30 / $16.50 |
+
+Probe results on 2026-09-24, as pools with `allow_fallbacks: true`:
+- 30 of 30 text and image requests returned HTTP 200.
+- OpenRouter routes by price, so `morph` serves most requests.
+
+Known limits:
+- `sail-research/us` serves GLM text only. OpenRouter skips it for image requests.
+- `coreweave/fp8` twice answered a DeepSeek image request as if no image was sent. It stays because it is the only other US-datacenter DeepSeek endpoint. A request reaches it only after Morph direct and Morph through OpenRouter both fail.
+- `coreweave/nvfp4` returns HTTP 429 from a shared pool most of the time; see below.
+- `decart/fp4` is fp4 quantization.
+
+Excluded on 2026-09-24:
+- **US headquarters without a confirmed US datacenter:** `wafer`, `together`, `parasail/*`, `io-net/fp8`, `novita/fp8`, `phala*`, `baseten/fp8`, `fireworks`, `deepinfra/*`, `modal*`, `inference-net/fp4`, `open-inference/fp4`, `crusoe/fp4`, `krea/fp8`, `sail-research/fp4`.
+- **Non-US or unknown location:** `z-ai/fp8` (SG), `siliconflow/fp8` (US datacenters, SG headquarters), `inceptron/fp8` (FI), `nextbit/fp8` (ES), `moonshotai/mxfp4` (SG), `dekallm` (ID), `relace`, `near-ai/fp8`, `digitalocean`, `reka`, `makora`.
+- **Image input rejected (HTTP 400):** `venice` (GLM) and `venice/fp8` (DeepSeek).
+- **Above `max_price`:** `morph/fast` for Kimi ($6.00 / $22.50).
+
+### Why GLM-5.3-Flash failed before 2026-09-24
+
+The route pinned one endpoint (`only: ['coreweave/nvfp4']`, `allow_fallbacks: false`). CoreWeave serves OpenRouter's non-BYOK traffic from a shared pool. On 2026-09-24 that pool returned HTTP 429 `rate_limit_exceeded` (`limit_source: upstream_provider_shared_pool`) for 11 of 15 requests that OpenRouter routed to it. With a single pin and no fallback, every such 429 reached the user. The same 429 was recorded on 2026-09-18.
+
+### OpenAI and Anthropic models are not managed
+
+The managed lineup offers open-weight models only. OpenAI and Anthropic models reach members through BYOK (`openai/<id>`, `anthropic/<id>`) or a ChatGPT plan (`codex/<id>`). They never bill Kortix credits. `src/managed.test.ts` fails when a managed entry routes to an `openai/` or `anthropic/` upstream. Claude Opus 5.5, GPT-6 Sol, and GPT-6 Luna were added as managed on 2026-09-24 (#7561) and removed the same day.
+
+`CATALOG` carries the models.dev records for `openai/gpt-6-sol`, `openai/gpt-6-luna`, `anthropic/claude-opus-5-5`, and their three OpenRouter ids. The BYOK and ChatGPT routes take reasoning effort, modalities, and `temperature` from them. The ChatGPT lineup (`apps/api/src/llm-gateway/models/codex-models.ts`) offers `codex/gpt-6-sol` and `codex/gpt-6-luna`.
+
+GLM-5.3 744B accepts only text input and is excluded.
 
 DeepSeek V4 Flash 0731 and DeepSeek V4 Pro 0813 remain excluded. DeepSeek reports that V4.1 Flash supersedes V4 Pro for performance, cost, speed, and task completion.
 
@@ -28,7 +74,7 @@ Qwen3.8 Max 0902 remains excluded. On 2026-09-21, OpenRouter listed one `alibaba
 
 ## Catalog
 
-`CATALOG` is the bundled models.dev snapshot. `MANAGED_MODELS` contains the managed lineup. `PLATFORM_DEFAULT_MODEL_ID` is `deepseek-v4.1-flash`. The runtime catalog refreshes from the configured models.dev URL.
+`CATALOG` is the bundled models.dev snapshot. It lives in `src/catalog-data.ts`, not in `index.ts`, so a bundler drops the ~7.6 MB JSON for consumers that never read `CATALOG` or `catalogModelForWireModel`. `MANAGED_MODELS` contains the managed lineup. `PLATFORM_DEFAULT_MODEL_ID` is `deepseek-v4.1-flash`. The runtime catalog refreshes from the configured models.dev URL.
 
 ## License
 

package/dist/catalog-data.d.ts

@@ -0,0 +1,30 @@
+import type { Catalog, CatalogModel } from './index.js';
+/**
+ * The bundled models.dev snapshot. Kept out of `index.ts` so consumers that
+ * never read it (browser bundles) do not ship it — see
+ * catalog-isolation.test.ts.
+ */
+export declare const CATALOG: Catalog;
+/**
+ * Resolve a gateway WIRE model id to its full `CatalogModel` — the capability
+ * record `generationControlCapabilities`/`clampGenerationConfig` gate against.
+ * The CANONICAL wire-id → model resolver: it stitches together the three id
+ * shapes a gateway request can carry, so a caller never has to string-split
+ * `<provider>/<model>` or special-case managed slugs itself.
+ *
+ *   - `codex/<id>`      → the genuine OpenAI catalog entry (`openai/<id>`).
+ *   - `<provider>/<id>` → that provider's catalog entry (BYOK models).
+ *   - bare `<id>`       → a managed slug, resolved via its `pricingRef` (the
+ *                         model's real models.dev id) so e.g. `claude-opus-4.8`
+ *                         gets Claude's real `reasoning_options`/`limit` instead
+ *                         of a permissive fallback; a synthesized minimal record
+ *                         (reasoning/tool_call/temperature all true) when the
+ *                         managed slug has no models.dev entry to borrow from.
+ *
+ * `catalog` defaults to the bundled static `CATALOG`. apps/api passes the LIVE
+ * models.dev snapshot instead (its own thin wrapper of the same name) so the
+ * host-side generation-controls clamp sees fresh capabilities; the standalone
+ * gateway transport, which has no runtime snapshot, uses the bundled catalog.
+ */
+export declare function catalogModelForWireModel(wireModel: string, catalog?: Catalog): CatalogModel | undefined;
+//# sourceMappingURL=catalog-data.d.ts.map

package/dist/catalog-data.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"catalog-data.d.ts","sourceRoot":"","sources":["../src/catalog-data.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,OAAO,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AAGrD;;;;GAIG;AACH,eAAO,MAAM,OAAO,EAAkB,OAAO,CAAC;AAU9C;;;;;;;;;;;;;;;;;;;;GAoBG;AACH,wBAAgB,wBAAwB,CACtC,SAAS,EAAE,MAAM,EACjB,OAAO,GAAE,OAAiB,GACzB,YAAY,GAAG,SAAS,CA6B1B"}

package/dist/catalog-data.js

@@ -0,0 +1,69 @@
+import catalogJson from './catalog.generated.json' with { type: 'json' };
+import { getManagedModel, pricingRefLookupCandidates } from './index.js';
+/**
+ * The bundled models.dev snapshot. Kept out of `index.ts` so consumers that
+ * never read it (browser bundles) do not ship it — see
+ * catalog-isolation.test.ts.
+ */
+export const CATALOG = catalogJson;
+function modelsByWireId(catalog) {
+    const byId = new Map();
+    for (const provider of catalog.providers) {
+        for (const model of provider.models)
+            byId.set(`${provider.id}/${model.id}`, model);
+    }
+    return byId;
+}
+/**
+ * Resolve a gateway WIRE model id to its full `CatalogModel` — the capability
+ * record `generationControlCapabilities`/`clampGenerationConfig` gate against.
+ * The CANONICAL wire-id → model resolver: it stitches together the three id
+ * shapes a gateway request can carry, so a caller never has to string-split
+ * `<provider>/<model>` or special-case managed slugs itself.
+ *
+ *   - `codex/<id>`      → the genuine OpenAI catalog entry (`openai/<id>`).
+ *   - `<provider>/<id>` → that provider's catalog entry (BYOK models).
+ *   - bare `<id>`       → a managed slug, resolved via its `pricingRef` (the
+ *                         model's real models.dev id) so e.g. `claude-opus-4.8`
+ *                         gets Claude's real `reasoning_options`/`limit` instead
+ *                         of a permissive fallback; a synthesized minimal record
+ *                         (reasoning/tool_call/temperature all true) when the
+ *                         managed slug has no models.dev entry to borrow from.
+ *
+ * `catalog` defaults to the bundled static `CATALOG`. apps/api passes the LIVE
+ * models.dev snapshot instead (its own thin wrapper of the same name) so the
+ * host-side generation-controls clamp sees fresh capabilities; the standalone
+ * gateway transport, which has no runtime snapshot, uses the bundled catalog.
+ */
+export function catalogModelForWireModel(wireModel, catalog = CATALOG) {
+    if (wireModel.startsWith('codex/')) {
+        return modelsByWireId(catalog).get(`openai/${wireModel.slice('codex/'.length)}`);
+    }
+    const slash = wireModel.indexOf('/');
+    if (slash > 0) {
+        const providerId = wireModel.slice(0, slash);
+        const modelId = wireModel.slice(slash + 1);
+        return catalog.providers
+            .find((provider) => provider.id === providerId)
+            ?.models.find((model) => model.id === modelId);
+    }
+    const managed = getManagedModel(wireModel);
+    if (managed) {
+        const catalogById = modelsByWireId(catalog);
+        const byPricingRef = pricingRefLookupCandidates(managed.pricingRef)
+            .map((ref) => catalogById.get(ref))
+            .find((entry) => entry !== undefined);
+        if (byPricingRef)
+            return byPricingRef;
+        return {
+            id: managed.id,
+            name: managed.name,
+            reasoning: true,
+            tool_call: true,
+            temperature: true,
+            limit: managed.limit,
+        };
+    }
+    return undefined;
+}
+//# sourceMappingURL=catalog-data.js.map

package/dist/catalog-data.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"catalog-data.js","sourceRoot":"","sources":["../src/catalog-data.ts"],"names":[],"mappings":"AAAA,OAAO,WAAW,MAAM,0BAA0B,CAAC,OAAO,IAAI,EAAE,MAAM,EAAE,CAAC;AAEzE,OAAO,EAAE,eAAe,EAAE,0BAA0B,EAAE,MAAM,SAAS,CAAC;AAEtE;;;;GAIG;AACH,MAAM,CAAC,MAAM,OAAO,GAAG,WAAsB,CAAC;AAE9C,SAAS,cAAc,CAAC,OAAgB;IACtC,MAAM,IAAI,GAAG,IAAI,GAAG,EAAwB,CAAC;IAC7C,KAAK,MAAM,QAAQ,IAAI,OAAO,CAAC,SAAS,EAAE,CAAC;QACzC,KAAK,MAAM,KAAK,IAAI,QAAQ,CAAC,MAAM;YAAE,IAAI,CAAC,GAAG,CAAC,GAAG,QAAQ,CAAC,EAAE,IAAI,KAAK,CAAC,EAAE,EAAE,EAAE,KAAK,CAAC,CAAC;IACrF,CAAC;IACD,OAAO,IAAI,CAAC;AACd,CAAC;AAED;;;;;;;;;;;;;;;;;;;;GAoBG;AACH,MAAM,UAAU,wBAAwB,CACtC,SAAiB,EACjB,UAAmB,OAAO;IAE1B,IAAI,SAAS,CAAC,UAAU,CAAC,QAAQ,CAAC,EAAE,CAAC;QACnC,OAAO,cAAc,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,UAAU,SAAS,CAAC,KAAK,CAAC,QAAQ,CAAC,MAAM,CAAC,EAAE,CAAC,CAAC;IACnF,CAAC;IACD,MAAM,KAAK,GAAG,SAAS,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;IACrC,IAAI,KAAK,GAAG,CAAC,EAAE,CAAC;QACd,MAAM,UAAU,GAAG,SAAS,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC;QAC7C,MAAM,OAAO,GAAG,SAAS,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC;QAC3C,OAAO,OAAO,CAAC,SAAS;aACrB,IAAI,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,QAAQ,CAAC,EAAE,KAAK,UAAU,CAAC;YAC/C,EAAE,MAAM,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,EAAE,KAAK,OAAO,CAAC,CAAC;IACnD,CAAC;IACD,MAAM,OAAO,GAAG,eAAe,CAAC,SAAS,CAAC,CAAC;IAC3C,IAAI,OAAO,EAAE,CAAC;QACZ,MAAM,WAAW,GAAG,cAAc,CAAC,OAAO,CAAC,CAAC;QAC5C,MAAM,YAAY,GAAG,0BAA0B,CAAC,OAAO,CAAC,UAAU,CAAC;aAChE,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,WAAW,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;aAClC,IAAI,CAAC,CAAC,KAAK,EAAyB,EAAE,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC;QAC/D,IAAI,YAAY;YAAE,OAAO,YAAY,CAAC;QACtC,OAAO;YACL,EAAE,EAAE,OAAO,CAAC,EAAE;YACd,IAAI,EAAE,OAAO,CAAC,IAAI;YAClB,SAAS,EAAE,IAAI;YACf,SAAS,EAAE,IAAI;YACf,WAAW,EAAE,IAAI;YACjB,KAAK,EAAE,OAAO,CAAC,KAAK;SACrB,CAAC;IACJ,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC"}

package/dist/catalog.generated.json

@@ -2,7 +2,7 @@
   "source": "https://models.dev/api.json",
   "fetched_at": "2026-08-25T16:21:02.770Z",
   "provider_count": 200,
-  "model_count": 7293,
+  "model_count": 7299,
   "providers": [
     {
       "id": "hpc-ai",
@@ -73198,6 +73198,187 @@
             }
           },
           "released": "2026-09-04"
+        },
+        {
+          "id": "openai/gpt-6-sol",
+          "name": "GPT-6 Sol",
+          "description": "GPT model for general reasoning, writing, coding, and tool-assisted tasks",
+          "released": "2026-09-22",
+          "attachment": true,
+          "reasoning": true,
+          "reasoning_options": [
+            {
+              "type": "effort",
+              "values": [
+                "none",
+                "low",
+                "medium",
+                "high",
+                "xhigh",
+                "max"
+              ]
+            }
+          ],
+          "tool_call": true,
+          "structured_output": true,
+          "open_weights": false,
+          "temperature": false,
+          "knowledge": "2026-04-20",
+          "last_updated": "2026-09-22",
+          "family": "gpt-sol",
+          "modalities": {
+            "input": [
+              "text",
+              "image",
+              "pdf"
+            ],
+            "output": [
+              "text"
+            ]
+          },
+          "limit": {
+            "context": 1050000,
+            "input": 922000,
+            "output": 128000
+          },
+          "cost": {
+            "input": 2,
+            "output": 10,
+            "cache_read": 0.2,
+            "cache_write": 2.5,
+            "tiers": [
+              {
+                "input": 4,
+                "output": 15,
+                "cache_read": 0.4,
+                "cache_write": 5,
+                "tier": {
+                  "type": "context",
+                  "size": 272000
+                }
+              }
+            ],
+            "context_over_200k": {
+              "input": 4,
+              "output": 15,
+              "cache_read": 0.4,
+              "cache_write": 5
+            }
+          }
+        },
+        {
+          "id": "openai/gpt-6-luna",
+          "name": "GPT-6 Luna",
+          "description": "GPT model for general reasoning, writing, coding, and tool-assisted tasks",
+          "released": "2026-09-22",
+          "attachment": true,
+          "reasoning": true,
+          "reasoning_options": [
+            {
+              "type": "effort",
+              "values": [
+                "none",
+                "low",
+                "medium",
+                "high",
+                "xhigh",
+                "max"
+              ]
+            }
+          ],
+          "tool_call": true,
+          "structured_output": true,
+          "open_weights": false,
+          "temperature": false,
+          "knowledge": "2026-05-18",
+          "last_updated": "2026-09-22",
+          "family": "gpt-luna",
+          "modalities": {
+            "input": [
+              "text",
+              "image",
+              "pdf"
+            ],
+            "output": [
+              "text"
+            ]
+          },
+          "limit": {
+            "context": 1050000,
+            "input": 922000,
+            "output": 128000
+          },
+          "cost": {
+            "input": 0.1,
+            "output": 0.5,
+            "cache_read": 0.01,
+            "cache_write": 0.125,
+            "tiers": [
+              {
+                "input": 0.2,
+                "output": 0.75,
+                "cache_read": 0.02,
+                "cache_write": 0.25,
+                "tier": {
+                  "type": "context",
+                  "size": 272000
+                }
+              }
+            ],
+            "context_over_200k": {
+              "input": 0.2,
+              "output": 0.75,
+              "cache_read": 0.02,
+              "cache_write": 0.25
+            }
+          }
+        },
+        {
+          "id": "anthropic/claude-opus-5.5",
+          "name": "Claude Opus 5.5",
+          "description": "Flagship Claude model for deep reasoning, coding, and long-horizon agents",
+          "released": "2026-09-22",
+          "attachment": true,
+          "reasoning": true,
+          "reasoning_options": [
+            {
+              "type": "effort",
+              "values": [
+                "low",
+                "medium",
+                "high",
+                "xhigh",
+                "max"
+              ]
+            }
+          ],
+          "tool_call": true,
+          "structured_output": true,
+          "open_weights": false,
+          "temperature": true,
+          "knowledge": "2026-06",
+          "last_updated": "2026-09-22",
+          "family": "claude-opus",
+          "modalities": {
+            "input": [
+              "text",
+              "image",
+              "pdf"
+            ],
+            "output": [
+              "text"
+            ]
+          },
+          "limit": {
+            "context": 1000000,
+            "output": 128000
+          },
+          "cost": {
+            "input": 4,
+            "output": 20,
+            "cache_read": 0.2,
+            "cache_write": 5
+          }
         }
       ]
     },
@@ -82519,6 +82700,53 @@
             "cache_read": 0.3,
             "cache_write": 3.75
           }
+        },
+        {
+          "id": "claude-opus-5-5",
+          "name": "Claude Opus 5.5",
+          "description": "Claude model for long-running agentic coding and knowledge work",
+          "released": "2026-09-22",
+          "attachment": true,
+          "reasoning": true,
+          "reasoning_options": [
+            {
+              "type": "effort",
+              "values": [
+                "low",
+                "medium",
+                "high",
+                "xhigh",
+                "max"
+              ]
+            }
+          ],
+          "tool_call": true,
+          "structured_output": true,
+          "open_weights": false,
+          "temperature": false,
+          "knowledge": "2026-06",
+          "last_updated": "2026-09-22",
+          "family": "claude-opus",
+          "modalities": {
+            "input": [
+              "text",
+              "image",
+              "pdf"
+            ],
+            "output": [
+              "text"
+            ]
+          },
+          "limit": {
+            "context": 1000000,
+            "output": 128000
+          },
+          "cost": {
+            "input": 4,
+            "output": 20,
+            "cache_read": 0.2,
+            "cache_write": 5
+          }
         }
       ]
     },
@@ -97410,6 +97638,140 @@
             }
           },
           "released": "2026-09-04"
+        },
+        {
+          "id": "gpt-6-sol",
+          "name": "GPT-6 Sol",
+          "description": "OpenAI model for complex coding and agentic workflows",
+          "released": "2026-09-22",
+          "attachment": true,
+          "reasoning": true,
+          "reasoning_options": [
+            {
+              "type": "effort",
+              "values": [
+                "none",
+                "low",
+                "medium",
+                "high",
+                "xhigh",
+                "max"
+              ]
+            }
+          ],
+          "tool_call": true,
+          "structured_output": true,
+          "open_weights": false,
+          "temperature": false,
+          "knowledge": "2026-04-20",
+          "last_updated": "2026-09-22",
+          "family": "gpt-sol",
+          "modalities": {
+            "input": [
+              "text",
+              "image",
+              "pdf"
+            ],
+            "output": [
+              "text"
+            ]
+          },
+          "limit": {
+            "context": 1050000,
+            "input": 922000,
+            "output": 128000
+          },
+          "cost": {
+            "input": 2,
+            "output": 10,
+            "cache_read": 0.2,
+            "cache_write": 2.5,
+            "tiers": [
+              {
+                "input": 4,
+                "output": 15,
+                "cache_read": 0.4,
+                "cache_write": 5,
+                "tier": {
+                  "type": "context",
+                  "size": 272000
+                }
+              }
+            ],
+            "context_over_200k": {
+              "input": 4,
+              "output": 15,
+              "cache_read": 0.4,
+              "cache_write": 5
+            }
+          }
+        },
+        {
+          "id": "gpt-6-luna",
+          "name": "GPT-6 Luna",
+          "description": "OpenAI's most efficient model for focused, high-volume tasks",
+          "released": "2026-09-22",
+          "attachment": true,
+          "reasoning": true,
+          "reasoning_options": [
+            {
+              "type": "effort",
+              "values": [
+                "none",
+                "low",
+                "medium",
+                "high",
+                "xhigh",
+                "max"
+              ]
+            }
+          ],
+          "tool_call": true,
+          "structured_output": true,
+          "open_weights": false,
+          "temperature": false,
+          "knowledge": "2026-05-18",
+          "last_updated": "2026-09-22",
+          "family": "gpt-luna",
+          "modalities": {
+            "input": [
+              "text",
+              "image",
+              "pdf"
+            ],
+            "output": [
+              "text"
+            ]
+          },
+          "limit": {
+            "context": 1050000,
+            "input": 922000,
+            "output": 128000
+          },
+          "cost": {
+            "input": 0.1,
+            "output": 0.5,
+            "cache_read": 0.01,
+            "cache_write": 0.125,
+            "tiers": [
+              {
+                "input": 0.2,
+                "output": 0.75,
+                "cache_read": 0.02,
+                "cache_write": 0.25,
+                "tier": {
+                  "type": "context",
+                  "size": 272000
+                }
+              }
+            ],
+            "context_over_200k": {
+              "input": 0.2,
+              "output": 0.75,
+              "cache_read": 0.02,
+              "cache_write": 0.25
+            }
+          }
         }
       ]
     },

package/dist/enablement.d.ts

@@ -59,7 +59,7 @@ export declare function defaultEnabledModelIds(models: EnablementCandidate[], op
  * id Bedrock refuses ("Invocation of model ID xai.grok-4.6 with on-demand
  * throughput isn't supported. Retry your request with the ID or ARN of an
  * inference profile"), then looped "Retrying in Ns" forever. Proven live on
- * the Essentia self-host 2026-08-26.
+ * the SampleCo self-host 2026-08-26.
  *
  * Rule: when a set carries ANY inference-profile id, its bare ids are not
  * auto-selectable. An inference profile is always invokable; a bare id may

package/dist/enablement.js

@@ -115,7 +115,7 @@ export function defaultEnabledModelIds(models, opts) {
  * id Bedrock refuses ("Invocation of model ID xai.grok-4.6 with on-demand
  * throughput isn't supported. Retry your request with the ID or ARN of an
  * inference profile"), then looped "Retrying in Ns" forever. Proven live on
- * the Essentia self-host 2026-08-26.
+ * the SampleCo self-host 2026-08-26.
  *
  * Rule: when a set carries ANY inference-profile id, its bare ids are not
  * auto-selectable. An inference profile is always invokable; a bare id may

package/dist/index.d.ts

@@ -159,12 +159,24 @@ export interface Catalog {
     model_count: number;
     providers: CatalogProvider[];
 }
-export declare const CATALOG: Catalog;
+export { CATALOG, catalogModelForWireModel } from './catalog-data.js';
+/**
+ * `{ id, env }` for every provider in `CATALOG`, in the same order — the only
+ * catalog fields `providerAuthRequirement` reads. ~8 KB instead of the ~7.6 MB
+ * snapshot, for browser code that needs provider credentials but no models.
+ * Regenerated together with the snapshot; catalog-isolation.test.ts checks
+ * that the two agree.
+ */
+export declare const CATALOG_PROVIDER_ENV: ReadonlyArray<{
+    id: string;
+    env: string[];
+}>;
 export interface ManagedModel {
     id: string;
     name: string;
     upstreamModelId: string;
     transport: 'openrouter';
+    morphModelId?: string;
     providerBrand?: string;
     pricingRef: string;
     pricing?: {
@@ -196,28 +208,6 @@ export declare const DEFAULT_MANAGED_MODEL_IDS: string[];
 export declare const MANAGED_FLAGSHIP_MODEL_ID: string;
 /** Concrete Kortix-managed default used when no account or project default exists. */
 export declare const PLATFORM_DEFAULT_MODEL_ID = "deepseek-v4.1-flash";
-/**
- * Resolve a gateway WIRE model id to its full `CatalogModel` — the capability
- * record `generationControlCapabilities`/`clampGenerationConfig` gate against.
- * The CANONICAL wire-id → model resolver: it stitches together the three id
- * shapes a gateway request can carry, so a caller never has to string-split
- * `<provider>/<model>` or special-case managed slugs itself.
- *
- *   - `codex/<id>`      → the genuine OpenAI catalog entry (`openai/<id>`).
- *   - `<provider>/<id>` → that provider's catalog entry (BYOK models).
- *   - bare `<id>`       → a managed slug, resolved via its `pricingRef` (the
- *                         model's real models.dev id) so e.g. `claude-opus-4.8`
- *                         gets Claude's real `reasoning_options`/`limit` instead
- *                         of a permissive fallback; a synthesized minimal record
- *                         (reasoning/tool_call/temperature all true) when the
- *                         managed slug has no models.dev entry to borrow from.
- *
- * `catalog` defaults to the bundled static `CATALOG`. apps/api passes the LIVE
- * models.dev snapshot instead (its own thin wrapper of the same name) so the
- * host-side generation-controls clamp sees fresh capabilities; the standalone
- * gateway transport, which has no runtime snapshot, uses the bundled catalog.
- */
-export declare function catalogModelForWireModel(wireModel: string, catalog?: Catalog): CatalogModel | undefined;
 export declare const MODEL_SELECTOR_PROVIDER_IDS: readonly ["kortix", "opencode", "anthropic", "openai", "github-copilot", "google", "openrouter", "vercel"];
 export declare const PROVIDER_LABELS: Record<string, string>;
 //# sourceMappingURL=index.d.ts.map