chore: refresh the generated catalog and limits tables

README.md

@@ -161,13 +161,13 @@ Zen's provider-level SDK is `@ai-sdk/openai-compatible`. models.dev names a **di
 
 | models.dev `provider.npm` | models | protocol | served from |
 | :-- | --: | :-- | :-- |
-| _(absent)_ | 24 | OpenAI Chat Completions | the route you configured |
+| _(absent)_ | 25 | OpenAI Chat Completions | the route you configured |
 | `@ai-sdk/openai` | 31 | OpenAI Responses | `opencode-responses` |
 | `@ai-sdk/anthropic` | 17 | Anthropic Messages | `opencode-anthropic` |
 | `@ai-sdk/google` | 8 | Google Generative AI | `opencode-google` |
 | `@ai-sdk/mistral` | 1 | _(a dialect this plugin does not serve)_ | **not offered** |
 
-Counts are the **81** active `opencode` models in `models.opencode.ai` as of 2026-11-03 — everything else the vendor lists is deprecated or retired, and the patch ships only what the gateway still serves. The same 81 are bundled offline in `src/catalog-data.ts`, so every one of them routes correctly before the first catalog refresh; regenerate that file with `pnpm run catalog:shim -- --write` (without `--write` it only checks).
+Counts are the **82** active `opencode` models in `models.opencode.ai` as of 2026-11-04 — everything else the vendor lists is deprecated or retired, and the patch ships only what the gateway still serves. The same 82 are bundled offline in `src/catalog-data.ts`, so every one of them routes correctly before the first catalog refresh; regenerate that file with `pnpm run catalog:shim -- --write` (without `--write` it only checks).
 
 The patch **keeps those two routes out of both the model picker and _Settings → Models_**. Three properties hold today, and one does not yet:
 

README.zh-CN.md

@@ -165,7 +165,7 @@ Zen 的 provider 级 SDK 是 `@ai-sdk/openai-compatible`。models.dev **只在
 | `@ai-sdk/google` | 8 | Google Generative AI | `opencode-google` |
 | `@ai-sdk/mistral` | 1 | _（本插件不服务的方言）_ | **不提供** |
 
-数量为 `models.opencode.ai` 中 `opencode` 的 81 个**在用**模型，统计于 2026-11-03 —— 其余已废弃或下架，补丁只内置网关仍在服务的部分。同样这 81 个模型离线内置在 `src/catalog-data.ts`，因此在第一次 catalog 刷新之前它们每一个路由都是正确的；用 `pnpm run catalog:shim -- --write` 重新生成该文件（不加 `--write` 时只做检查）。
+数量为 `models.opencode.ai` 中 `opencode` 的 82 个**在用**模型，统计于 2026-11-04 —— 其余已废弃或下架，补丁只内置网关仍在服务的部分。同样这 82 个模型离线内置在 `src/catalog-data.ts`，因此在第一次 catalog 刷新之前它们每一个路由都是正确的；用 `pnpm run catalog:shim -- --write` 重新生成该文件（不加 `--write` 时只做检查）。
 
 插件把这两条内部路由同时挡在模型选择器和 _Settings → Models_ 之外。**三条性质今天就成立，还有一条尚未成立：**
 

docs/protocol-routing-and-merge.md

@@ -8,11 +8,11 @@ OpenCode serves its models on **four different API shapes**, and which one a mod
 
 | API shape | Wire endpoint | Which models | Signal in the catalog |
 | :-- | :-- | :-- | :-- |
-| OpenAI **Chat Completions** | `POST /zen/v1/chat/completions` | **the default**: 24 of 81 | the model names **no** SDK |
-| OpenAI **Responses** | `POST /zen/v1/responses` | 31 of 81 | `@ai-sdk/openai` |
-| Anthropic **Messages** | `POST /zen/v1/messages` | 17 of 81 | `@ai-sdk/anthropic` |
-| Google **Generative AI** | `POST /zen/v1/models/<id>:streamGenerateContent` | 8 of 81 | `@ai-sdk/google` |
-| Mistral **Conversations** | `POST /zen/v1/chat/completions` | 1 of 81 | `@ai-sdk/mistral` |
+| OpenAI **Chat Completions** | `POST /zen/v1/chat/completions` | **the default**: 25 of 82 | the model names **no** SDK |
+| OpenAI **Responses** | `POST /zen/v1/responses` | 31 of 82 | `@ai-sdk/openai` |
+| Anthropic **Messages** | `POST /zen/v1/messages` | 17 of 82 | `@ai-sdk/anthropic` |
+| Google **Generative AI** | `POST /zen/v1/models/<id>:streamGenerateContent` | 8 of 82 | `@ai-sdk/google` |
+| Mistral **Conversations** | `POST /zen/v1/chat/completions` | 1 of 82 | `@ai-sdk/mistral` |
 
 Counts are the shipped catalog. They are asserted against this table by `catalog.test.ts`, because a count written into prose does not regenerate with the data it counts.
 

docs/redirect-planes.md

@@ -10,7 +10,7 @@ A **plane** is one gateway base URL with its own catalog and its own credential.
 
 | Plane | Route id | Base URL | Credential | Models |
 | :-- | :-- | :-- | :-- | --: |
-| Zen (pay-as-you-go, free tier) | `opencode` | the `endpoint` family's | `OPENCODE_API_KEY` | 81 served / 131 listed |
+| Zen (pay-as-you-go, free tier) | `opencode` | the `endpoint` family's | `OPENCODE_API_KEY` | 82 served / 132 listed |
 | Go (subscription) | `opencode-go` | the `endpoint` family's | `OPENCODE_GO_API_KEY` | 30 served / 37 listed |
 
 Counts as of 2026-11-03 (the shipped shim, regenerated in #132 and refreshed for `pulse-alpha-free`); the document's 2026-10-09 stamp is for its measurements.
@@ -27,13 +27,13 @@ A **shape** is one wire protocol, named by the SDK a model declares in the catal
 
 | SDK in the catalog  | Protocol                | Zen |    Go |
 | :------------------ | :---------------------- | --: | ----: |
-| _(none)_            | `openai-completions`    |  24 |    19 |
+| _(none)_            | `openai-completions`    |  25 |    19 |
 | `@ai-sdk/openai`    | `openai-responses`      |  31 | **5** |
 | `@ai-sdk/anthropic` | `anthropic-messages`    |  17 | **5** |
 | `@ai-sdk/mistral`   | `mistral-conversations` |   1 |     — |
 | `@ai-sdk/google`    | `google-generative-ai`  |   8 | **1** |
 
-**Absence is the signal, and it is a real one.** Both planes set a provider-level `npm` of `@ai-sdk/openai-compatible` in `models.dev`, and **no model declares it explicitly**: 0 of 81 on Zen, 0 of 30 on Go. A model that speaks the completions shape simply omits `provider.npm` and inherits it. So "no SDK" does not mean "unknown"; it means _the provider default_, which is the protocol the route was itself registered with.
+**Absence is the signal, and it is a real one.** Both planes set a provider-level `npm` of `@ai-sdk/openai-compatible` in `models.dev`, and **no model declares it explicitly**: 0 of 82 on Zen, 0 of 30 on Go. A model that speaks the completions shape simply omits `provider.npm` and inherits it. So "no SDK" does not mean "unknown"; it means _the provider default_, which is the protocol the route was itself registered with.
 
 That is why the servability check treats `undefined` as servable, and why `@ai-sdk/openai-compatible` is deliberately **not** in `PROTOCOL_FOR_SDK`: listing it would record the default as though it were an override. The coverage guard still covers the case where a model declares it explicitly. None do today, and if one did, the test would name it rather than drop the model.
 

src/catalog-data.ts

@@ -1,3 +1,3 @@
 /**
  * Static OpenCode Go & Zen model catalog data.
  *
@@ -161,7 +161,7 @@
     id: "deepseek-v4-pro",
     input_modalities: ["text"],
     max_output_tokens: 384_000,
-    name: "DeepSeek V4 Pro (New)",
+    name: "DeepSeek V4 Pro",
     reasoning: true,
     reasoning_options: [{ type: "effort", values: ["high", "max"] }],
   },
@@ -869,6 +869,26 @@
       { type: "effort", values: ["low", "medium", "high", "xhigh", "max"] },
     ],
   },
+  {
+    context_window: 1_048_576,
+    cost: {
+      cache_read: 0,
+      input: 0,
+      output: 0,
+    },
+    id: "crouton-alpha-free",
+    input_modalities: ["text", "image"],
+    is_free: true,
+    max_output_tokens: 270_000,
+    name: "Crouton Alpha Free",
+    reasoning: true,
+    reasoning_options: [
+      {
+        type: "effort",
+        values: ["minimal", "low", "medium", "high", "xhigh", "max"],
+      },
+    ],
+  },
   {
     context_window: 1_000_000,
     cost: {

test/catalog.test.ts

@@ -198,12 +198,12 @@ describe("OpenCode Model Catalog & Enrichment", () => {
     // replacing the whole set — so these counts are the cold-start picker, and a
     // silent drop here is a model nobody can pick. Regenerate with
     // `scripts/regenerate-catalog-shim.ts` rather than editing by hand.
-    expect(OPENCODE_ZEN_CATALOG.length).toBe(81);
+    expect(OPENCODE_ZEN_CATALOG.length).toBe(82);
     const freeModels = OPENCODE_ZEN_CATALOG.filter((m) => m.is_free === true);
     // Only the free tiers the registry still lists as active. The count is the
     // `is_free` flag the shim carries, NOT the `-free` suffix: `big-pickle` is
     // flagged and is not suffixed, so a suffix rule would have missed it.
-    expect(freeModels.length).toBe(9);
+    expect(freeModels.length).toBe(10);
 
     const ids = new Set(OPENCODE_ZEN_CATALOG.map((m) => m.id));
     expect(ids.has("beam-free")).toBe(true);