AI preset model refresh + thinking-probe fix + 0.51.0-beta.1

package.json

@@ -1,6 +1,6 @@
 {
   "name": "open-alice",
-  "version": "0.50.0-beta.1",
+  "version": "0.51.0-beta.1",
   "description": "File-based trading agent engine",
   "type": "module",
   "main": "dist/electron/main.js",

src/ai-providers/preset-catalog.ts

@@ -73,7 +73,6 @@ export const CLAUDE_OAUTH: PresetDef = {
   }),
   models: [
     { id: 'claude-opus-4-8', label: 'Claude Opus 4.8' },
-    { id: 'claude-opus-4-7', label: 'Claude Opus 4.7' },
     { id: 'claude-sonnet-4-6', label: 'Claude Sonnet 4.6' },
   ],
 }
@@ -84,7 +83,7 @@ export const CLAUDE_API: PresetDef = {
   description: 'Pay per token via Anthropic API',
   category: 'official',
   defaultName: 'Claude (API Key)',
-  hint: 'Model is switchable here or from the profile list anytime. Opus is ~5× the cost of Sonnet; Haiku is cheapest for high-volume work.',
+  hint: 'Model is switchable here or from the profile list anytime. Opus is ~5× the cost of Sonnet.',
   zodSchema: z.object({
     backend: z.literal('agent-sdk'),
     loginMethod: z.literal('api-key'),
@@ -93,9 +92,7 @@ export const CLAUDE_API: PresetDef = {
   }),
   models: [
     { id: 'claude-opus-4-8', label: 'Claude Opus 4.8' },
-    { id: 'claude-opus-4-7', label: 'Claude Opus 4.7' },
     { id: 'claude-sonnet-4-6', label: 'Claude Sonnet 4.6' },
-    { id: 'claude-haiku-4-5', label: 'Claude Haiku 4.5' },
   ],
   regions: [{ id: 'official', label: 'Official (api.anthropic.com)', wires: { anthropic: '' } }],
   writeOnlyFields: ['apiKey'],
@@ -118,7 +115,6 @@ export const CODEX_OAUTH: PresetDef = {
   models: [
     { id: 'gpt-5.5', label: 'GPT 5.5' },
     { id: 'gpt-5.4', label: 'GPT 5.4' },
-    { id: 'gpt-5.4-mini', label: 'GPT 5.4 Mini' },
   ],
 }
 
@@ -137,7 +133,6 @@ export const CODEX_API: PresetDef = {
   models: [
     { id: 'gpt-5.5', label: 'GPT 5.5' },
     { id: 'gpt-5.4', label: 'GPT 5.4' },
-    { id: 'gpt-5.4-mini', label: 'GPT 5.4 Mini' },
   ],
   // Same key + base; the shape is how you call it. Responses is OpenAI's
   // current API (what codex speaks); Chat Completions is the legacy shape
@@ -215,12 +210,12 @@ export const GLM: PresetDef = {
   description: 'Zhipu GLM models via Claude Agent SDK (Anthropic-compatible)',
   category: 'third-party',
   defaultName: 'GLM',
-  hint: 'China console: bigmodel.cn — International console: z.ai. API keys are region-locked. GLM 5.1 is the current flagship, served on both regions.',
+  hint: 'China console: bigmodel.cn — International console: z.ai. API keys are region-locked. GLM 5.2 is the current flagship, served on both regions.',
   zodSchema: z.object({
     backend: z.literal('agent-sdk'),
     loginMethod: z.literal('api-key'),
     baseUrl: z.string().default('https://open.bigmodel.cn/api/anthropic').describe('API endpoint'),
-    model: z.string().default('glm-5.1').describe('Model'),
+    model: z.string().default('glm-5.2').describe('Model'),
     apiKey: z.string().min(1).describe('GLM API key'),
   }),
   regions: [
@@ -232,9 +227,7 @@ export const GLM: PresetDef = {
     } },
   ],
   models: [
-    { id: 'glm-5.1', label: 'GLM 5.1' },
-    { id: 'glm-4.7', label: 'GLM 4.7' },
-    { id: 'glm-4.5-air', label: 'GLM 4.5 Air' },
+    { id: 'glm-5.2', label: 'GLM 5.2' },
   ],
   writeOnlyFields: ['apiKey'],
 }
@@ -257,7 +250,7 @@ export const KIMI: PresetDef = {
     backend: z.literal('agent-sdk'),
     loginMethod: z.literal('api-key'),
     baseUrl: z.string().default('https://api.moonshot.cn/anthropic').describe('API endpoint'),
-    model: z.string().default('kimi-k2.6').describe('Model'),
+    model: z.string().default('kimi-k2.7-code').describe('Model'),
     apiKey: z.string().min(1).describe('Moonshot API key'),
   }),
   regions: [
@@ -269,8 +262,8 @@ export const KIMI: PresetDef = {
     } },
   ],
   models: [
+    { id: 'kimi-k2.7-code', label: 'Kimi K2.7 Code' },
     { id: 'kimi-k2.6', label: 'Kimi K2.6' },
-    { id: 'kimi-k2.5', label: 'Kimi K2.5' },
   ],
   writeOnlyFields: ['apiKey'],
 }
@@ -298,7 +291,6 @@ export const DEEPSEEK: PresetDef = {
   ],
   models: [
     { id: 'deepseek-v4-pro', label: 'DeepSeek V4 Pro (flagship)' },
-    { id: 'deepseek-v4-flash', label: 'DeepSeek V4 Flash (cheap/fast)' },
   ],
   writeOnlyFields: ['apiKey'],
 }

src/workspaces/agent-probe.ts

@@ -41,26 +41,51 @@ export interface CodexProbeInput {
   wireApi: 'chat' | 'responses';
 }
 
+/** Does this error mean the model MANDATES extended thinking? Some reasoning
+ *  models (e.g. Kimi k2.7) 400 with "invalid thinking: only type=enabled is
+ *  allowed for this model" when the request omits thinking — they can't run it
+ *  disabled. We detect that to retry with thinking on, rather than enabling it
+ *  for every model (Claude/GLM/etc. don't need it and some reject the param). */
+function isThinkingRequiredError(err: unknown): boolean {
+  const e = err as { status?: number; message?: string } | undefined;
+  return e?.status === 400 && typeof e?.message === 'string' && /thinking/i.test(e.message);
+}
+
 export async function probeAnthropic(input: ClaudeProbeInput): Promise<ProbeResult> {
   // `authToken` makes the SDK send `Authorization: Bearer`; `apiKey` makes it
   // send `x-api-key`. Pick exactly one — sending both can trip gateways that
   // reject ambiguous auth, and Anthropic's own API now 401s OAuth-via-Bearer.
   const client = input.authMode === 'bearer'
     ? new Anthropic({ authToken: input.apiKey, baseURL: input.baseUrl })
     : new Anthropic({ apiKey: input.apiKey, baseURL: input.baseUrl });
-  const msg = await client.messages.create({
-    model: input.model,
-    // Enough room for a reasoning model to finish thinking AND emit a visible
-    // reply on a trivial prompt — a tiny budget gets spent entirely on reasoning,
-    // leaving empty content (the "(empty reply)" the user saw). One-off per Test.
-    max_tokens: 512,
-    messages: [{ role: 'user', content: 'Hi' }],
-  });
-  const text = msg.content
+
+  const extract = (msg: Anthropic.Message): string => msg.content
     .filter((b): b is Anthropic.TextBlock => b.type === 'text')
     .map((b) => b.text)
     .join('');
-  return { text };
+
+  try {
+    const msg = await client.messages.create({
+      model: input.model,
+      // Enough room for a reasoning model to finish thinking AND emit a visible
+      // reply on a trivial prompt — a tiny budget gets spent entirely on reasoning,
+      // leaving empty content (the "(empty reply)" the user saw). One-off per Test.
+      max_tokens: 512,
+      messages: [{ role: 'user', content: 'Hi' }],
+    });
+    return { text: extract(msg) };
+  } catch (err) {
+    if (!isThinkingRequiredError(err)) throw err;
+    // Thinking-mandatory model: retry with it enabled. budget_tokens must be
+    // < max_tokens and Anthropic's floor is 1024, so bump max_tokens to match.
+    const msg = await client.messages.create({
+      model: input.model,
+      max_tokens: 2048,
+      thinking: { type: 'enabled', budget_tokens: 1024 },
+      messages: [{ role: 'user', content: 'Hi' }],
+    });
+    return { text: extract(msg) };
+  }
 }
 
 export async function probeOpenAI(input: CodexProbeInput): Promise<ProbeResult> {

ui/src/components/workspace/WorkspaceAIConfigModal.tsx

@@ -463,7 +463,7 @@ export function WorkspaceAIConfigModal({ wsId, onClose }: Props) {
               value={form.model}
               suggestions={modelSuggestions}
               onChange={(v) => setForm({ ...form, model: v })}
-              placeholder={tab === 'claude' ? 'claude-sonnet-4-6' : tab === 'opencode' || tab === 'pi' ? 'deepseek-chat' : 'gpt-4o'}
+              placeholder={tab === 'claude' ? 'claude-opus-4-8' : tab === 'opencode' || tab === 'pi' ? 'deepseek-chat' : 'gpt-5.5'}
             />
             {modelSuggestions.length > 0 && (
               <p className="text-[11px] text-text-muted/70 mt-1">Suggestions from the matched provider — or type any model id.</p>