fix: retire claude-3-haiku-20240307 and gpt-4o (closes #44)

CHANGELOG.md

@@ -3,6 +3,20 @@
 All notable changes to `@stackbilt/llm-providers` are documented here.
 Format follows [Keep a Changelog](https://keepachangelog.com/). Versions use [Semantic Versioning](https://semver.org/).
 
+## [1.4.0] — 2026-04-17
+
+### Deprecated
+- **`MODELS.CLAUDE_3_HAIKU`** (`claude-3-haiku-20240307`) — Anthropic retires 2026-04-19. Migrate to `MODELS.CLAUDE_HAIKU_4_5` or `MODELS.CLAUDE_3_5_HAIKU`. Export retained; callers get a compile-time `@deprecated` warning.
+- **`MODELS.GPT_4O`** (`gpt-4o`) — retired by OpenAI on 2026-04-03. Migrate to `MODELS.GPT_4O_MINI` or a current GPT-4 successor. Export retained; callers get a compile-time `@deprecated` warning.
+
+### Removed
+- `claude-3-haiku-20240307` — dropped from `AnthropicProvider.models[]` and its capabilities/pricing table. Calls to this ID will fail at Anthropic's cutoff; keeping it advertised would mislead consumers.
+- `gpt-4o` — dropped from `OpenAIProvider.models[]` and its capabilities/pricing table.
+- `gpt-4-turbo-preview` — dead alias dropped from `OpenAIProvider.models[]` (no corresponding capabilities entry; caught by the new drift test).
+
+### Added
+- **Model drift test** (`src/__tests__/model-drift.test.ts`) — asserts every provider's `models[]` array is symmetrically covered by its capabilities map. Prevents future retirement drift where a model is removed from one list but not the other. Runs across all 5 providers.
+
 ## [1.3.0] — 2026-04-16
 
 ### Added

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@stackbilt/llm-providers",
-  "version": "1.3.0",
+  "version": "1.4.0",
   "description": "Multi-LLM failover with circuit breakers, cost tracking, and intelligent retry. Cloudflare Workers native.",
   "author": "Stackbilt <admin@stackbilt.dev>",
   "license": "Apache-2.0",

src/__tests__/factory.test.ts

@@ -49,22 +49,22 @@ const mockOpenAIProvider = {
 
 const mockAnthropicProvider = {
   name: 'anthropic',
-  models: ['claude-3-haiku-20240307', 'claude-3-sonnet-20240229'],
+  models: ['claude-haiku-4-5-20251001', 'claude-3-sonnet-20240229'],
   supportsStreaming: true,
   supportsTools: true,
   supportsBatching: false,
   supportsVision: true,
   generateResponse: vi.fn().mockResolvedValue({
     message: 'Anthropic response',
     usage: { inputTokens: 10, outputTokens: 20, totalTokens: 30, cost: 0.002 },
-    model: 'claude-3-haiku-20240307',
+    model: 'claude-haiku-4-5-20251001',
     provider: 'anthropic',
     responseTime: 1200
   } as LLMResponse),
   streamResponse: vi.fn(),
   getProviderBalance: vi.fn(),
   validateConfig: vi.fn().mockReturnValue(true),
-  getModels: vi.fn().mockReturnValue(['claude-3-haiku-20240307', 'claude-3-sonnet-20240229']),
+  getModels: vi.fn().mockReturnValue(['claude-haiku-4-5-20251001', 'claude-3-sonnet-20240229']),
   estimateCost: vi.fn().mockReturnValue(0.002),
   healthCheck: vi.fn().mockResolvedValue(true),
   getMetrics: vi.fn().mockReturnValue({
@@ -161,7 +161,7 @@ describe('LLMProviderFactory', () => {
     mockAnthropicProvider.generateResponse.mockReset().mockResolvedValue({
       message: 'Anthropic response',
       usage: { inputTokens: 10, outputTokens: 20, totalTokens: 30, cost: 0.002 },
-      model: 'claude-3-haiku-20240307',
+      model: 'claude-haiku-4-5-20251001',
       provider: 'anthropic',
       responseTime: 1200
     } as LLMResponse);
@@ -326,7 +326,7 @@ describe('LLMProviderFactory', () => {
       // Test Claude model request
       const claudeRequest: LLMRequest = {
         ...testRequest,
-        model: 'claude-3-haiku-20240307'
+        model: 'claude-haiku-4-5-20251001'
       };
 
       await factory.generateResponse(claudeRequest);
@@ -373,7 +373,7 @@ describe('LLMProviderFactory', () => {
         fallbackRules: [{
           condition: 'error',
           fallbackProvider: 'anthropic',
-          fallbackModel: 'claude-3-haiku-20240307'
+          fallbackModel: 'claude-haiku-4-5-20251001'
         }]
       });
 
@@ -385,7 +385,7 @@ describe('LLMProviderFactory', () => {
       });
 
       expect(mockAnthropicProvider.generateResponse).toHaveBeenCalledWith(
-        expect.objectContaining({ model: 'claude-3-haiku-20240307' })
+        expect.objectContaining({ model: 'claude-haiku-4-5-20251001' })
       );
     });
   });

src/__tests__/from-env.test.ts

@@ -13,13 +13,13 @@ const mockValidateConfig = vi.fn().mockReturnValue(true);
 vi.mock('../providers/openai', () => ({
   OpenAIProvider: vi.fn().mockImplementation(() => ({
     name: 'openai',
-    models: ['gpt-4o'],
+    models: ['gpt-4o-mini'],
     supportsStreaming: true,
     supportsTools: true,
     supportsBatching: true,
     validateConfig: mockValidateConfig,
     generateResponse: vi.fn(),
-    getModels: vi.fn().mockReturnValue(['gpt-4o']),
+    getModels: vi.fn().mockReturnValue(['gpt-4o-mini']),
     estimateCost: vi.fn().mockReturnValue(0),
     healthCheck: vi.fn().mockResolvedValue(true),
     getMetrics: vi.fn().mockReturnValue({
@@ -33,13 +33,13 @@ vi.mock('../providers/openai', () => ({
 vi.mock('../providers/anthropic', () => ({
   AnthropicProvider: vi.fn().mockImplementation(() => ({
     name: 'anthropic',
-    models: ['claude-3-haiku-20240307'],
+    models: ['claude-haiku-4-5-20251001'],
     supportsStreaming: true,
     supportsTools: true,
     supportsBatching: false,
     validateConfig: mockValidateConfig,
     generateResponse: vi.fn(),
-    getModels: vi.fn().mockReturnValue(['claude-3-haiku-20240307']),
+    getModels: vi.fn().mockReturnValue(['claude-haiku-4-5-20251001']),
     estimateCost: vi.fn().mockReturnValue(0),
     healthCheck: vi.fn().mockResolvedValue(true),
     getMetrics: vi.fn().mockReturnValue({

src/__tests__/model-drift.test.ts

@@ -0,0 +1,43 @@
+/**
+ * Model drift tests
+ *
+ * Guards against retired/unknown model IDs slipping through by verifying
+ * that every entry in a provider's `models` array has a matching entry in
+ * its capabilities map. A mismatch usually means a model was retired from
+ * one list but not the other — catch that at CI time, not at runtime.
+ */
+
+import { describe, it, expect } from 'vitest';
+import { AnthropicProvider } from '../providers/anthropic';
+import { OpenAIProvider } from '../providers/openai';
+import { CloudflareProvider } from '../providers/cloudflare';
+import { CerebrasProvider } from '../providers/cerebras';
+import { GroqProvider } from '../providers/groq';
+
+type WithCapabilities = {
+  models: string[];
+  getModelCapabilities: () => Record<string, unknown>;
+};
+
+const providers: Array<[string, WithCapabilities]> = [
+  ['anthropic', new AnthropicProvider({ apiKey: 'test' }) as unknown as WithCapabilities],
+  ['openai', new OpenAIProvider({ apiKey: 'test' }) as unknown as WithCapabilities],
+  ['cloudflare', new CloudflareProvider({ ai: { run: async () => ({}) } as never, accountId: 'test' }) as unknown as WithCapabilities],
+  ['cerebras', new CerebrasProvider({ apiKey: 'test' }) as unknown as WithCapabilities],
+  ['groq', new GroqProvider({ apiKey: 'test' }) as unknown as WithCapabilities]
+];
+
+describe('model drift', () => {
+  it.each(providers)('%s: every advertised model has a capabilities entry', (_name, provider) => {
+    const caps = provider.getModelCapabilities();
+    const missing = provider.models.filter((m) => !(m in caps));
+    expect(missing).toEqual([]);
+  });
+
+  it.each(providers)('%s: every capabilities entry is advertised in models[]', (_name, provider) => {
+    const caps = provider.getModelCapabilities();
+    const advertised = new Set(provider.models);
+    const orphaned = Object.keys(caps).filter((m) => !advertised.has(m));
+    expect(orphaned).toEqual([]);
+  });
+});

src/__tests__/response-format.test.ts

@@ -47,11 +47,11 @@ describe('OpenAI response_format', () => {
   });
 
   it('should pass response_format through to the API body', async () => {
-    mockFetch.mockResolvedValueOnce(openAIChatCompletion('gpt-4o'));
+    mockFetch.mockResolvedValueOnce(openAIChatCompletion('gpt-4o-mini'));
 
     await provider.generateResponse({
       messages: [{ role: 'user', content: 'Give me JSON' }],
-      model: 'gpt-4o',
+      model: 'gpt-4o-mini',
       response_format: { type: 'json_object' }
     });
 
@@ -60,23 +60,23 @@ describe('OpenAI response_format', () => {
   });
 
   it('should not include response_format when not set', async () => {
-    mockFetch.mockResolvedValueOnce(openAIChatCompletion('gpt-4o'));
+    mockFetch.mockResolvedValueOnce(openAIChatCompletion('gpt-4o-mini'));
 
     await provider.generateResponse({
       messages: [{ role: 'user', content: 'Hello' }],
-      model: 'gpt-4o'
+      model: 'gpt-4o-mini'
     });
 
     const body = JSON.parse(mockFetch.mock.calls[0][1].body);
     expect(body.response_format).toBeUndefined();
   });
 
   it('should pass response_format with type text', async () => {
-    mockFetch.mockResolvedValueOnce(openAIChatCompletion('gpt-4o', 'plain text'));
+    mockFetch.mockResolvedValueOnce(openAIChatCompletion('gpt-4o-mini', 'plain text'));
 
     await provider.generateResponse({
       messages: [{ role: 'user', content: 'Hello' }],
-      model: 'gpt-4o',
+      model: 'gpt-4o-mini',
       response_format: { type: 'text' }
     });