weave-router: cheap5 (5-model cost-aware, oracle 83.7 offline)

Author: steventohme

router_inference/config/weave-router.json

@@ -7,11 +7,10 @@
       "deepseek/deepseek-v4-pro",
       "gemini-3.1-flash-lite-preview",
       "gemini-3.1-pro-preview",
-      "qwen/qwen3-235b-a22b-2507",
-      "claude-sonnet-4-5"
+      "qwen/qwen3-235b-a22b-2507"
     ],
-    "description": "Weave Router (oracle 6-model trimmed pool, margin=0.05): drop low-coverage models",
-    "router_version": "oracle-6m",
+    "description": "Weave Router (cheap5): 5-model cost-aware oracle, margin=0.05",
+    "router_version": "cheap5",
     "router_homepage": "https://workweave.ai"
   }
 }

router_inference/predictions/weave-router-robustness.json

@@ -968,7 +968,7 @@
   {
     "global index": "MMLUPro_economics_6114",
     "prompt": "Kindly examine the subsequent multiple-choice inquiries and select the option that most plausibly represents the correct choice based on the given alternatives.\n\nContext: None\n\nQuestion: Analyze the following equation to identify the type of model it most accurately characterizes:\n\n\\[ y_{{it}} = \\alpha + \\beta_{{it}} + \\mu_i + \\nu_{{it}} \\]\n\nChoices: \nA. A strictly temporal series model\nB. A model of difference-in-differences\nC. A model with random effects\nD. A model using pooled ordinary least squares\nE. A model involving mixed effects\nF. A model with spatial autocorrelation\nG. A model of quantile regression\nH. A model of panel data\nI. A model with entity fixed effects\nJ. A model with time fixed effects\n\nIndicate the correct option by using \\(\\boxed{X}\\), where \\(X\\) corresponds to the accurate choice. Construct the explanation or feedback in no more than three sentences.",
-    "prediction": "claude-sonnet-4-5",
+    "prediction": "gemini-3.1-flash-lite-preview",
     "cost": 0.0,
     "for_optimality": false
   },
@@ -1640,7 +1640,7 @@
   {
     "global index": "MathQA_202",
     "prompt": "Carefully examine the subsequent multiple-choice queries and determine the option that most plausibly represents the correct response from those provided.\n\nContext: None\n\nQuestion: Is the mean value between two collections of numbers nearer to the one containing a greater quantity of numbers?\n\nOptions:\nA. 70\nB. 80\nC. 85\nD. 90\nE. 95\n\nPresent the correct letter choice in \\boxed{X}, where X signifies the accurate option. Limit the explanation or feedback to no more than three sentences.",
-    "prediction": "claude-sonnet-4-5",
+    "prediction": "gemini-3.1-flash-lite-preview",
     "cost": 0.0,
     "for_optimality": false
   },
@@ -2809,7 +2809,7 @@
   {
     "global index": "WMT19-de-en_46",
     "prompt": "Translt the folowing scentnce from Englsih to Grman.\n\nTe trvel waring is also a responce to a nw Misouri lw that woud mak it mor difficult to sue a buisness for housign or employmnt discrimnation.\n\nProide yor final anser in \\boxed{} format.",
-    "prediction": "qwen/qwen3-235b-a22b-2507",
+    "prediction": "deepseek/deepseek-v4-flash",
     "cost": 0.0,
     "for_optimality": false
   },
@@ -2823,7 +2823,7 @@
   {
     "global index": "WMT19-de-en_883",
     "prompt": "Pladis, the company that owns McVitie's, announced it has poured over £5 million into the location during recent years.",
-    "prediction": "qwen/qwen3-235b-a22b-2507",
+    "prediction": "gemini-3.1-flash-lite-preview",
     "cost": 0.0,
     "for_optimality": false
   },
@@ -2837,7 +2837,7 @@
   {
     "global index": "WMT19-fi-en_610",
     "prompt": "Trnslate teh folowing sentnce from Englesh to Finnish.\n\nOn Satrday, rains r expecetd in mnay prts of souther and cetnral arreas with a chnce of thundrstorms.\n\nProvdie your fnal answre in \\boxed{{}} frmat.",
-    "prediction": "claude-sonnet-4-5",
+    "prediction": "qwen/qwen3-235b-a22b-2507",
     "cost": 0.0,
     "for_optimality": false
   },