@blockrun/clawrouter 0.12.48 → 0.12.50

package/dist/cli.js

@@ -1649,12 +1649,14 @@ var DEFAULT_ROUTING_CONFIG = {
   tiers: {
     SIMPLE: {
       primary: "google/gemini-2.5-flash",
-      // 1,238ms, 60% retention (best) — fast AND quality
+      // 1,238ms, IQ 20, 60% retention (best) — fast AND quality
       fallback: [
+        "google/gemini-3-flash-preview",
+        // 1,398ms, IQ 46 — smarter fallback
         "deepseek/deepseek-chat",
-        // 1,431ms, 41% retention
+        // 1,431ms, IQ 32, 41% retention
         "moonshot/kimi-k2.5",
-        // 1,646ms, strong quality
+        // 1,646ms, IQ 47, strong quality
         "google/gemini-2.5-flash-lite",
         // 1,353ms, 1M context, ultra cheap ($0.10/$0.40)
         "xai/grok-4-fast-non-reasoning",
@@ -1665,10 +1667,12 @@ var DEFAULT_ROUTING_CONFIG = {
     },
     MEDIUM: {
       primary: "moonshot/kimi-k2.5",
-      // 1,646ms, $0.60/$3.00 — strong tool use, quality output
+      // 1,646ms, IQ 47, $0.60/$3.00 — strong tool use, quality output
       fallback: [
+        "google/gemini-3-flash-preview",
+        // 1,398ms, IQ 46 — nearly same IQ, faster + cheaper
         "deepseek/deepseek-chat",
-        // 1,431ms, 41% retention
+        // 1,431ms, IQ 32, 41% retention
         "google/gemini-2.5-flash",
         // 1,238ms, 60% retention
         "google/gemini-2.5-flash-lite",
@@ -1681,24 +1685,24 @@ var DEFAULT_ROUTING_CONFIG = {
     },
     COMPLEX: {
       primary: "google/gemini-3.1-pro",
-      // 1,609ms — fast flagship quality
+      // 1,609ms, IQ 57 — fast flagship quality
       fallback: [
-        "google/gemini-2.5-flash",
-        // 1,238ms, cheap failsafe before expensive models
-        "google/gemini-2.5-flash-lite",
-        // 1,353ms, 1M context, ultra-cheap failsafe ($0.10/$0.40)
         "google/gemini-3-pro-preview",
-        // 1,352ms
+        // 1,352ms, IQ 48 — quality-first fallback
+        "google/gemini-3-flash-preview",
+        // 1,398ms, IQ 46 — fast + smart
+        "xai/grok-4-0709",
+        // 1,348ms, IQ 41
         "google/gemini-2.5-pro",
         // 1,294ms
-        "xai/grok-4-0709",
-        // 1,348ms
-        "deepseek/deepseek-chat",
-        // 1,431ms
         "anthropic/claude-sonnet-4.6",
-        // 2,110ms — quality fallback
+        // 2,110ms, IQ 52 — quality fallback
+        "deepseek/deepseek-chat",
+        // 1,431ms, IQ 32
+        "google/gemini-2.5-flash",
+        // 1,238ms, IQ 20 — cheap last resort
         "openai/gpt-5.4"
-        // 6,213ms — slowest but highest quality
+        // 6,213ms, IQ 57 — slowest but highest quality
       ]
     },
     REASONING: {
@@ -5587,6 +5591,27 @@ function normalizeMessagesForThinking(messages) {
   });
   return hasChanges ? normalized : messages;
 }
+function debrandSystemMessages(messages, resolvedModel) {
+  const PROFILE_NAMES = ["auto", "free", "eco", "premium"];
+  const profilePattern = new RegExp(
+    `\\bblockrun/(${PROFILE_NAMES.join("|")})\\b`,
+    "gi"
+  );
+  const prefixPattern = /\bblockrun\/(?=[a-z])/gi;
+  let hasChanges = false;
+  const result = messages.map((msg) => {
+    if (msg.role !== "system" || typeof msg.content !== "string") return msg;
+    let content = msg.content;
+    const afterProfiles = content.replace(profilePattern, resolvedModel);
+    const afterPrefix = afterProfiles.replace(prefixPattern, "");
+    if (afterPrefix !== content) {
+      hasChanges = true;
+      content = afterPrefix;
+    }
+    return content !== msg.content ? { ...msg, content } : msg;
+  });
+  return hasChanges ? result : messages;
+}
 function truncateMessages(messages) {
   if (!messages || messages.length <= MAX_MESSAGES) {
     return {
@@ -6351,6 +6376,12 @@ async function tryModelRequest(upstreamUrl, method, headers, body, modelId, maxT
     if (Array.isArray(parsed.messages)) {
       parsed.messages = normalizeMessageRoles(parsed.messages);
     }
+    if (Array.isArray(parsed.messages)) {
+      parsed.messages = debrandSystemMessages(
+        parsed.messages,
+        modelId
+      );
+    }
     if (Array.isArray(parsed.messages)) {
       const truncationResult = truncateMessages(parsed.messages);
       parsed.messages = truncationResult.messages;