@sayknow-cli/ai 0.5.26 → 0.6.0

package/CHANGELOG.md

@@ -2,6 +2,11 @@
 
 ## [Unreleased]
 
+## [0.6.0] - 2026-09-24
+### Fixed
+- Retire `anthropic/claude-3-haiku-20240307`, `claude-3-5-sonnet-20240620` and `claude-3-5-sonnet-20241022`: the Anthropic API answers 404 `not_found_error` for all three (measured 2026-09-23 with a forced tool call). Left in the catalog, the cheapest of them was the first pick for typed decisions on every registry with an Anthropic key.
+- The on-disk model cache now filters retired keys on read. A row written before a retirement carried the dead entry for up to the TTL and the catalog merge brought it straight back.
+
 ## [0.5.26] - 2026-09-23
 ### Fixed
 - API-key `/login` now upserts a second key instead of replacing the whole provider pool. Same-key re-login still reuses the existing row.

package/dist/types/model-retirements.d.ts

@@ -1,4 +1,4 @@
-export declare const RETIRED_MODEL_KEYS: readonly ["google-antigravity/gemini-3.1-pro-high"];
+export declare const RETIRED_MODEL_KEYS: readonly ["google-antigravity/gemini-3.1-pro-high", "anthropic/claude-3-haiku-20240307", "anthropic/claude-3-5-sonnet-20240620", "anthropic/claude-3-5-sonnet-20241022"];
 export declare function isRetiredModelKey(provider: string, modelId: string): boolean;
 export declare function isRetiredModel(model: {
     provider: string;

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@sayknow-cli/ai",
-	"version": "0.5.26",
+	"version": "0.6.0",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"homepage": "https://sayknow-cli.com",
 	"author": "jaybeyond",
@@ -43,7 +43,7 @@
 	"dependencies": {
 		"@anthropic-ai/sdk": "^0.94.0",
 		"@bufbuild/protobuf": "^2.12.0",
-		"@sayknow-cli/utils": "0.5.26",
+		"@sayknow-cli/utils": "0.6.0",
 		"openai": "^6.36.0",
 		"partial-json": "^0.1.7",
 		"zod": "4.4.3"

package/src/model-cache.ts

@@ -4,6 +4,7 @@
  */
 import { Database } from "bun:sqlite";
 import { getModelDbPath } from "@sayknow-cli/utils/dirs";
+import { isRetiredModelKey } from "./model-retirements";
 import type { Api, Model } from "./types";
 
 const CACHE_SCHEMA_VERSION = 3;
@@ -96,7 +97,12 @@ export function readModelCache<TApi extends Api>(
 		if (!row || row.version !== CACHE_SCHEMA_VERSION) {
 			return null;
 		}
-		const models = JSON.parse(row.models) as Model<TApi>[];
+		// A row written before a model was retired still carries it for up to the TTL;
+		// the bundled catalog filters retirements, so the cache must too or the merge
+		// brings the dead entry straight back.
+		const models = (JSON.parse(row.models) as Model<TApi>[]).filter(
+			model => !isRetiredModelKey(model.provider ?? providerId, model.id),
+		);
 		const ageMs = now() - row.updated_at;
 		const fresh = Number.isFinite(ageMs) && ageMs >= 0 && ageMs <= ttlMs;
 		return {

package/src/model-retirements.ts

@@ -1,6 +1,18 @@
-// Retired from advertised catalogs because Cloud Code Assist rejects live calls
-// with HTTP 400. The callable high-thinking path is gemini-3.1-pro-low:high.
-export const RETIRED_MODEL_KEYS = ["google-antigravity/gemini-3.1-pro-high"] as const;
+// Retired from advertised catalogs because the provider rejects live calls.
+//
+// google-antigravity/gemini-3.1-pro-high: Cloud Code Assist answers HTTP 400. The
+// callable high-thinking path is gemini-3.1-pro-low:high.
+//
+// anthropic/claude-3-*: the Anthropic API answers 404 `not_found_error` for all three
+// (measured 2026-09-23 with a forced tool call). Left in the catalog, the cheapest of
+// them ($0.25/MTok) was the first pick for typed decisions, so the "small hosted model"
+// fallback silently answered nothing on every registry that had an Anthropic key.
+export const RETIRED_MODEL_KEYS = [
+	"google-antigravity/gemini-3.1-pro-high",
+	"anthropic/claude-3-haiku-20240307",
+	"anthropic/claude-3-5-sonnet-20240620",
+	"anthropic/claude-3-5-sonnet-20241022",
+] as const;
 
 const RETIRED_MODEL_KEY_SET = new Set<string>(RETIRED_MODEL_KEYS);