fix: strip reasoning/thinking params for models that don't support them (#766)

Models like antigravity/claude-sonnet-4-6 route through Google's internal
Cloud Code API which returns HTTP 400 when thinking/reasoning parameters
are included in the request body.

Changes:
- open-sse/services/modelCapabilities.ts: add supportsReasoning() function
  with a denylist of known-unsupported patterns (antigravity/claude-sonnet-*)
  and a registry-based lookup hook (supportsReasoning flag per model)
- open-sse/services/thinkingBudget.ts: in applyThinkingBudget(), add early
  exit before the mode switch — if model string is present and
  supportsReasoning() returns false, call stripThinkingConfig() immediately
  regardless of the configured ThinkingMode

This is fully backward-compatible: models not in the denylist are unaffected,
and the supportsReasoning registry flag defaults to null (pass-through).

Fixes: HTTP 400 errors on antigravity provider when client sends requests
with thinking/reasoning budget parameters (e.g. claude-sonnet-4-6 via AG).

Co-authored-by: oyi77 <oyi77@github.com>
Co-authored-by: oyi77 <oyi77@users.noreply.github.com>

open-sse/services/modelCapabilities.ts

@@ -48,3 +48,54 @@ export function supportsToolCalling(modelStr: string): boolean {
 
   return !blocked;
 }
+
+// Models that do NOT support reasoning/thinking parameters.
+// AG (Antigravity) claude-sonnet-4-6 routes through a Google internal API
+// that returns 400 if thinking params are included.
+const REASONING_UNSUPPORTED_PATTERNS = [
+  "antigravity/claude-sonnet-4-6",
+  "antigravity/claude-sonnet-4-5",
+  "antigravity/claude-sonnet-4",
+  "ag/claude-sonnet-4-6",
+  "ag/claude-sonnet-4-5",
+  "ag/claude-sonnet-4",
+];
+
+function getRegistryReasoningFlag(providerIdOrAlias: string, modelId: string): boolean | null {
+  const providerAlias = PROVIDER_ID_TO_ALIAS[providerIdOrAlias] || providerIdOrAlias;
+  const models = PROVIDER_MODELS[providerAlias];
+  if (!Array.isArray(models)) return null;
+  const found = models.find((m) => m?.id === modelId);
+  if (!found) return null;
+  return typeof found.supportsReasoning === "boolean" ? found.supportsReasoning : null;
+}
+
+/**
+ * Returns whether a model supports reasoning/thinking parameters.
+ *
+ * Decision order:
+ * 1) Provider registry metadata (supportsReasoning flag) when available.
+ * 2) Explicit denylist for known unsupported models (e.g. AG Claude Sonnet).
+ * 3) Default true (pass through — safe, provider will ignore if unsupported).
+ */
+export function supportsReasoning(modelStr: string): boolean {
+  const parsed = parseModel(modelStr);
+  const provider = parsed.provider || parsed.providerAlias || "";
+  const model = parsed.model || modelStr;
+
+  if (provider) {
+    const fromRegistry = getRegistryReasoningFlag(provider, model);
+    if (fromRegistry !== null) return fromRegistry;
+  }
+
+  const normalized = String(modelStr || "").toLowerCase();
+  if (!normalized) return true;
+
+  const blocked = REASONING_UNSUPPORTED_PATTERNS.some((pattern) =>
+    normalized === pattern ||
+    normalized.endsWith(`/${pattern}`) ||
+    normalized.includes(pattern)
+  );
+
+  return !blocked;
+}

open-sse/services/thinkingBudget.ts

@@ -14,6 +14,7 @@ export const ThinkingMode = {
 };
 
 import { capThinkingBudget, getDefaultThinkingBudget } from "@/shared/constants/modelSpecs";
+import { supportsReasoning } from "./modelCapabilities.ts";
 
 // Effort → budget token mapping
 export const EFFORT_BUDGETS = {
@@ -151,6 +152,13 @@ export function applyThinkingBudget(body, config = null) {
   const cfg = config || _config;
   if (!body || typeof body !== "object") return body;
 
+  // Early exit: strip ALL reasoning/thinking params for models that don't support them.
+  // Sending thinking params to unsupported models (e.g. AG claude-sonnet-4-6) causes 400 errors.
+  const modelStr = typeof body.model === "string" ? body.model : "";
+  if (modelStr && !supportsReasoning(modelStr)) {
+    return stripThinkingConfig(body);
+  }
+
   // Pre-processing: convert string thinkingLevel to numeric budget
   let processed = normalizeThinkingLevel(body);