pi-coralbricks-provider 1.0.25 → 1.0.26

package/README.md

@@ -2,7 +2,7 @@
 
 # 🪸 pi-coralbricks-provider
 
-**GLM 5.3, GLM 5.3 Flash & DeepSeek V4.1 Flash through [Coral Bricks](https://www.coralbricks.ai)**
+**GLM 5.3 & DeepSeek V4.1 Flash (FP4 variants included) through [Coral Bricks](https://www.coralbricks.ai)**
 
 _A [pi](https://github.com/earendil-works/pi-coding-agent) provider extension for Coral's OpenAI-compatible inference gateway — up to **1M context** on open models._
 
@@ -22,8 +22,8 @@ Run `bun run test:pi` for offline manifest, catalog, lifecycle and streaming che
 
 ## Features
 
-- **3 reasoning models** from Coral's live catalog — GLM 5.3 FP4, GLM 5.3 Flash, and DeepSeek V4.1 Flash
-- **1M token context** on GLM and DeepSeek, with vision (image input) on GLM 5.3 Flash and DeepSeek V4.1 Flash
+- **4 reasoning models** from Coral's live catalog — GLM 5.3 and its FP4 variant, DeepSeek V4.1 Flash and its FP4 variant
+- **1M token context** on GLM and DeepSeek, with vision (image input) on DeepSeek V4.1 Flash and its FP4 variant
 - **OpenAI-compatible API** — standard `/v1/chat/completions`, streaming, and tool calling
 - **Per-family thinking levels** — zai-style `thinking` control for GLM (including a *real* off switch), `reasoning_effort` for DeepSeek V4.1 Flash
 - **Accurate cost tracking** — input, cache-write and output rates mirror Coral's [published pricing](https://www.coralbricks.ai/pricing), and cached reads are **$0 on every model**
@@ -86,7 +86,9 @@ pi
 | Model | Context | Vision | Reasoning | Input $/M | Cache Read $/M | Cache Write $/M | Output $/M |
 |-------|---------|--------|-----------|-----------|-----------------|------------------|------------|
 | DeepSeek V4.1 Flash | 1.0M | ✅ | ✅ | $0.30 | — | $0.09 | $1.20 |
+| DeepSeek V4.1 Flash FP4 | 1.0M | ✅ | ✅ | $0.30 | — | $0.09 | $1.20 |
 | GLM 5.3 | 1.0M | ❌ | ✅ | $1.12 | — | $1.68 | $4.40 |
+| GLM 5.3 FP4 | 1.0M | ❌ | ✅ | $1.12 | — | $1.68 | $4.40 |
 
 *Costs are per million tokens. Cache Read shows — because Coral bills cached input at **$0** on every model. Prompt tokens Coral has not cached yet are billed once at the Cache Write rate, in place of the Input rate. Prices subject to change — check [Coral's live catalog](https://www.coralbricks.ai/api/public/models).*
 
@@ -168,7 +170,7 @@ Verified against the live gateway:
 | Model | Format | off | low | medium | high | max |
 |-------|--------|-----|-----|--------|------|-----|
 | GLM 5.3 FP4 | `thinking: {type}` + `reasoning_effort` | ✅ | ✅ | — | ✅ | ✅ |
-| GLM 5.3 Flash | `thinking: {type}` + `reasoning_effort` | ✅ | ✅ | — | ✅ | ✅ |
+| GLM 5.3 Flash (retired) | `thinking: {type}` + `reasoning_effort` | ✅ | ✅ | — | ✅ | ✅ |
 | DeepSeek V4.1 Flash | `reasoning_effort` | ✅ | ✅ | — | ✅ | ✅ |
 
 - **GLM** accepts zai-style `thinking: {type: "disabled"}` on Coral — pi's *off* level turns thinking off, though a short preamble of a few dozen tokens can still appear (the upstream Z.ai API has no off at all, so this differs from the canonical Z.ai map).
@@ -186,7 +188,7 @@ Coral's gateway follows the OpenAI Chat Completions API:
 
 ### Patch Overrides & Custom Models
 
-- **`patch.json`** — per-model overrides applied on top of `models.json` (reasoning flags, pricing corrections, compat settings, thinking level maps). Currently carries two entries: GLM 5.3 Flash (reasoning, image input, its thinking level map and compat settings) and DeepSeek V4.1 Flash (reasoning, its thinking level map and compat settings).
+- **`patch.json`** — per-model overrides applied on top of `models.json` (reasoning flags, pricing corrections, compat settings, thinking level maps). Currently one entry per live model: GLM 5.3 and its FP4 twin (reasoning, thinking level map, zai-style compat settings) and DeepSeek V4.1 Flash and its FP4 twin (reasoning, image input, thinking level map, `openai`-style compat settings).
 - **`custom-models.json`** — full model definitions for models Coral doesn't list. Merged after patch.
 
 Merge order: `[live|cache|embedded] → patch.json → custom-models.json`
@@ -198,7 +200,7 @@ Validated with [synbad](https://github.com/synthetic-lab/synbad) — Synthetic's
 | Model | Unary | Stream | Notes |
 |-------|-------|--------|-------|
 | GLM 5.3 FP4 | 13/13 ✅ | 13/13 ✅ | 5/5 runs |
-| GLM 5.3 Flash | 12/13 ⚠️ | 12/13 ⚠️ | `reasoning/reasoning-parsing` passes 4 of 5 runs; see below |
+| GLM 5.3 Flash (retired) | 12/13 ⚠️ | 12/13 ⚠️ | `reasoning/reasoning-parsing` passes 4 of 5 runs; see below |
 | GPT-OSS 120B (retired) | 12/13 ⚠️ | 12/13 ⚠️ | `tools/parallel-tool` 0 of 5 runs, before Coral retired the model |
 | DeepSeek V4.1 Flash | 12/13 ⚠️ | 12/13 ⚠️ | `tools/octo-list-no-optional-args` passes 1 of 5 runs; see below |
 | Kimi K3 (retired) | 13/13 ✅ | 13/13 ✅ | 5/5 runs, before Coral retired the model |

package/deprecated-models.json

@@ -1,103 +1 @@
-{
-  "deepseek-v4.1-flash-fast-fp4": {
-    "id": "deepseek-v4.1-flash-fast-fp4",
-    "name": "DeepSeek V4.1 Flash",
-    "reasoning": false,
-    "input": [
-      "text",
-      "image"
-    ],
-    "cost": {
-      "input": 0.3,
-      "output": 1.2,
-      "cacheRead": 0,
-      "cacheWrite": 0.09
-    },
-    "contextWindow": 1048576,
-    "maxTokens": 32768,
-    "compat": {
-      "supportsDeveloperRole": false,
-      "supportsStore": false,
-      "maxTokensField": "max_tokens"
-    },
-    "deprecatedAt": "2026-10-05T00:00:07.034Z"
-  },
-  "glm-5.3-flash-fp4": {
-    "id": "glm-5.3-flash-fp4",
-    "name": "GLM 5.3 Flash",
-    "reasoning": false,
-    "input": [
-      "text",
-      "image"
-    ],
-    "cost": {
-      "input": 0.15,
-      "output": 0.5,
-      "cacheRead": 0,
-      "cacheWrite": 0.23
-    },
-    "contextWindow": 1048576,
-    "maxTokens": 32768,
-    "compat": {
-      "supportsDeveloperRole": false,
-      "supportsStore": false,
-      "maxTokensField": "max_tokens"
-    },
-    "deprecatedAt": "2026-10-05T00:00:07.034Z"
-  },
-  "glm-5.3-fp4": {
-    "id": "glm-5.3-fp4",
-    "name": "GLM 5.3 FP4",
-    "reasoning": true,
-    "input": [
-      "text"
-    ],
-    "cost": {
-      "input": 1.12,
-      "output": 4.4,
-      "cacheRead": 0,
-      "cacheWrite": 1.68
-    },
-    "contextWindow": 1048576,
-    "maxTokens": 131072,
-    "thinkingLevelMap": {
-      "off": "none",
-      "minimal": null,
-      "low": "low",
-      "medium": null,
-      "high": "high",
-      "xhigh": null,
-      "max": "max"
-    },
-    "compat": {
-      "supportsStore": false,
-      "supportsDeveloperRole": false,
-      "supportsReasoningEffort": true,
-      "maxTokensField": "max_tokens",
-      "thinkingFormat": "zai"
-    },
-    "deprecatedAt": "2026-10-05T00:00:07.034Z"
-  },
-  "glm-5.3-flash-fast": {
-    "id": "glm-5.3-flash-fast",
-    "name": "GLM 5.3 Flash",
-    "reasoning": false,
-    "input": [
-      "text"
-    ],
-    "cost": {
-      "input": 0.15,
-      "output": 0.5,
-      "cacheRead": 0,
-      "cacheWrite": 0
-    },
-    "contextWindow": 1048576,
-    "maxTokens": 32768,
-    "compat": {
-      "supportsDeveloperRole": false,
-      "supportsStore": false,
-      "maxTokensField": "max_tokens"
-    },
-    "deprecatedAt": "2026-10-07T00:00:09.219Z"
-  }
-}
+{}

package/models.json

@@ -4,13 +4,36 @@
     "name": "DeepSeek V4.1 Flash",
     "reasoning": false,
     "input": [
-      "text"
+      "text",
+      "image"
+    ],
+    "cost": {
+      "input": 0.3,
+      "output": 1.2,
+      "cacheRead": 0,
+      "cacheWrite": 0.09
+    },
+    "contextWindow": 1048576,
+    "maxTokens": 32768,
+    "compat": {
+      "supportsDeveloperRole": false,
+      "supportsStore": false,
+      "maxTokensField": "max_tokens"
+    }
+  },
+  {
+    "id": "deepseek-v4.1-flash-fast-fp4",
+    "name": "Deepseek V4.1 Flash Fast Fp4",
+    "reasoning": false,
+    "input": [
+      "text",
+      "image"
     ],
     "cost": {
       "input": 0.3,
       "output": 1.2,
       "cacheRead": 0,
-      "cacheWrite": 0
+      "cacheWrite": 0.09
     },
     "contextWindow": 1048576,
     "maxTokens": 32768,
@@ -31,7 +54,28 @@
       "input": 1.12,
       "output": 4.4,
       "cacheRead": 0,
-      "cacheWrite": 0
+      "cacheWrite": 1.68
+    },
+    "contextWindow": 1048576,
+    "maxTokens": 32768,
+    "compat": {
+      "supportsDeveloperRole": false,
+      "supportsStore": false,
+      "maxTokensField": "max_tokens"
+    }
+  },
+  {
+    "id": "glm-5.3-fp4",
+    "name": "Glm 5.3 Fp4",
+    "reasoning": false,
+    "input": [
+      "text"
+    ],
+    "cost": {
+      "input": 1.12,
+      "output": 4.4,
+      "cacheRead": 0,
+      "cacheWrite": 1.68
     },
     "contextWindow": 1048576,
     "maxTokens": 32768,

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "pi-coralbricks-provider",
-  "version": "1.0.25",
-  "description": "CoralBricks provider extension for pi - Access GLM 5.3, GLM 5.3 Flash, and DeepSeek V4.1 Flash through the Coral Inference API",
+  "version": "1.0.26",
+  "description": "CoralBricks provider extension for pi - Access GLM 5.3 and DeepSeek V4.1 Flash (FP4 variants included) through the Coral Inference API",
   "author": "monotykamary",
   "homepage": "https://github.com/monotykamary/pi-coralbricks-provider#readme",
   "repository": {

package/patch.json

@@ -7,6 +7,7 @@
     "compat": { "supportsReasoningEffort": true, "thinkingFormat": "openai" }
   },
   "deepseek-v4.1-flash-fast-fp4": {
+    "name": "DeepSeek V4.1 Flash FP4",
     "reasoning": true,
     "input": ["text", "image"],
     "thinkingLevelMap": { "off": "none", "minimal": null, "low": "low", "medium": null, "high": "high", "xhigh": null, "max": "max" },
@@ -21,26 +22,11 @@
     "compat": { "supportsReasoningEffort": true, "thinkingFormat": "zai" }
   },
   "glm-5.3-fp4": {
+    "name": "GLM 5.3 FP4",
     "reasoning": true,
     "thinkingLevelMap": { "off": "none", "minimal": null, "low": "low", "medium": null, "high": "high", "xhigh": null, "max": "max" },
     "maxTokens": 131072,
     "cost": { "cacheWrite": 1.68 },
     "compat": { "supportsReasoningEffort": true, "thinkingFormat": "zai" }
-  },
-  "glm-5.3-flash-fast": {
-    "reasoning": true,
-    "input": ["text", "image"],
-    "thinkingLevelMap": { "off": "none", "minimal": null, "low": "low", "medium": null, "high": "high", "xhigh": null, "max": "max" },
-    "maxTokens": 131072,
-    "cost": { "cacheWrite": 0.23 },
-    "compat": { "supportsReasoningEffort": true, "thinkingFormat": "zai" }
-  },
-  "glm-5.3-flash-fp4": {
-    "reasoning": true,
-    "input": ["text", "image"],
-    "thinkingLevelMap": { "off": "none", "minimal": null, "low": "low", "medium": null, "high": "high", "xhigh": null, "max": "max" },
-    "maxTokens": 131072,
-    "cost": { "cacheWrite": 0.23 },
-    "compat": { "supportsReasoningEffort": true, "thinkingFormat": "zai" }
   }
 }

package/scripts/update-models.js

@@ -14,6 +14,10 @@
  *    same control-plane data without a key (contextWindow like "1M", inputPerM,
  *    outputPerM). Used when no key is configured or /v1/models fails.
  *
+ * The catalog is a subset: it lists each model's canonical slug, not every
+ * variant enabled for a key, so a catalog run only ever adds or keeps models —
+ * it never delists one (see keepUnlistedModels).
+ *
  * models.json is the source of truth for curated specs — the script preserves
  * existing data and only adds new models with API-derived defaults.
  * Curate models.json manually after new model discovery.
@@ -219,18 +223,38 @@ async function fetchPublicCatalog() {
   return models;
 }
 
-async function fetchModels() {
-  const apiKey = resolveApiKey();
+async function fetchModels(apiKey) {
   if (apiKey) {
     try {
-      return await fetchV1Models(apiKey);
+      return { models: await fetchV1Models(apiKey), authoritative: true };
     } catch (error) {
       console.warn(`⚠ /v1/models failed (${error.message}); falling back to public catalog`);
     }
   } else {
     console.warn('⚠ No API key found (no `coralbricks` credential in ' + AUTH_JSON_PATH + ', CORALBRICKS_API_KEY unset); using the public catalog');
   }
-  return fetchPublicCatalog();
+  return { models: await fetchPublicCatalog(), authoritative: false };
+}
+
+/**
+ * The public catalog is a subset of the per-key /v1/models set: it advertises a
+ * model's canonical slug, not every variant enabled for the key (the -fp4 twins
+ * are listed by /v1/models only). It can vouch for what exists, never for what
+ * is gone, so a non-authoritative fetch keeps every already-known model it does
+ * not list: a keyless run may discover models, it may not delist them. Returns
+ * the ids it kept, in the order they were appended.
+ */
+function keepUnlistedModels(models, existingModelsMap, authoritative) {
+  if (authoritative) return [];
+  const seen = new Set(models.map((m) => m.id));
+  const kept = [];
+  for (const [id, existing] of Object.entries(existingModelsMap)) {
+    if (!seen.has(id)) {
+      models.push(existing);
+      kept.push(id);
+    }
+  }
+  return kept;
 }
 
 // Transform API model → models.json entry
@@ -547,7 +571,8 @@ async function main() {
       return;
     }
 
-    const apiModels = await fetchModels();
+    const apiKey = resolveApiKey();
+    const { models: apiModels, authoritative } = await fetchModels(apiKey);
 
     // Load existing models.json — source of truth for curated specs
     const existingModels = loadJson(MODELS_JSON_PATH);
@@ -562,15 +587,20 @@ async function main() {
       isCatalogShape ? transformCatalogModel(m, existingModelsMap) : transformApiModel(m, existingModelsMap)
     );
 
-    // Live API is authoritative — models absent from API are removed
-    // (embedded data is already used for enrichment in transformApiModel)
+    // Live API is authoritative — models absent from it are removed (embedded
+    // data is already used for enrichment in transformApiModel). A catalog run
+    // is not: keep every known model it does not list.
+    const kept = keepUnlistedModels(models, existingModelsMap, authoritative);
+    if (kept.length > 0) {
+      console.warn(`⚠ Public catalog listed ${kept.length} fewer model(s) than models.json; kept: ${kept.join(', ')}`);
+    }
 
     // Sort by model name
     models.sort((a, b) => a.name.localeCompare(b.name));
 
     // Save models.json (pure API output, no patch/custom baked in)
     // Move delisted models to deprecated-models.json BEFORE models.json is overwritten
-    await updateDeprecatedModels(MODELS_JSON_PATH, models, resolveApiKey());
+    await updateDeprecatedModels(MODELS_JSON_PATH, models, apiKey);
     saveJson(MODELS_JSON_PATH, models);
 
     // Build the README model list: base → patch → custom. Grace-period models stay
@@ -608,4 +638,4 @@ if (process.argv[1] && path.resolve(process.argv[1]) === path.resolve(fileURLToP
   main();
 }
 
-export { isRetiredUpstream, updateDeprecatedModels };
+export { isRetiredUpstream, keepUnlistedModels, updateDeprecatedModels };