diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
index 11ca43d8..ea2fb245 100644
--- a/.pre-commit-config.yaml
+++ b/.pre-commit-config.yaml
@@ -66,6 +66,13 @@ repos:
       entry: tools/addlicense.sh
       pass_filenames: true
       files: '\.py$'
+    - id: model-cost-in-sync
+      name: model_cost.json is generated from model_profiles.yaml
+      language: python
+      entry: python scripts/pricing/build_model_cost.py --check
+      additional_dependencies: [pyyaml]
+      pass_filenames: false
+      files: '^(model_cost/|scripts/pricing/)'
 # Add mypy hooks for both local runs and CI (manual stage)
 - repo: local
   hooks:
diff --git a/llm_evaluation/evaluate_models.py b/llm_evaluation/evaluate_models.py
index 854dd1f8..36d4bd3e 100644
--- a/llm_evaluation/evaluate_models.py
+++ b/llm_evaluation/evaluate_models.py
@@ -16,6 +16,7 @@
 import json
 import argparse
 import glob
+import threading
 from typing import Dict, List, Any, Optional
 import sys
 
@@ -87,6 +88,10 @@ def __init__(
             str, Any
         ] = {}  # Store existing results for incremental evaluation
         self.cost_config: Dict[str, Any] = {}  # Store cost configuration
+        # Model names charged at the maximum price because no price profile
+        # matched them -> number of rows.
+        self.unpriced_models: Dict[str, int] = {}
+        self._unpriced_lock = threading.Lock()
         self.num_workers = num_workers
 
         # Load dataset configurations
@@ -207,21 +212,39 @@ def load_cost_config(self):
             self.cost_config = {}
 
     def _lookup_cost_info(self, model_name: str):
-        """Find the pricing entry for a model name, trying an exact match first
-        and then a substring fallback (historical behaviour). Returns the cost
-        dict, or None if no price is known."""
+        """Find the pricing entry for a model name by exact match on a model id
+        or alias (see model_cost/model_profiles.yaml). Returns the cost dict, or
+        None if no price is known.
+
+        There is deliberately no fuzzy/substring fallback: it used to let a name
+        pick up whichever price key happened to be a substring of it."""
         if not self.cost_config or not model_name:
             return None
         # Remove _batch suffix if present for cost lookup
         cost_lookup_name = (
             model_name[:-6] if model_name.endswith("_batch") else model_name
         )
-        if cost_lookup_name in self.cost_config:
-            return self.cost_config[cost_lookup_name]
-        for config_name in self.cost_config.keys():
-            if config_name in cost_lookup_name or cost_lookup_name in config_name:
-                return self.cost_config[config_name]
-        return None
+        return self.cost_config.get(cost_lookup_name)
+
+    def max_price_info(self) -> Dict[str, float]:
+        """The highest input and output prices in the cost table. Used to charge
+        rows whose model has no price profile, so an unregistered or misspelled
+        model is never cheaper than a registered one."""
+        if not self.cost_config:
+            return {
+                "input_token_price_per_million": 0.0,
+                "output_token_price_per_million": 0.0,
+            }
+        return {
+            "input_token_price_per_million": max(
+                float(v.get("input_token_price_per_million", 0.0))
+                for v in self.cost_config.values()
+            ),
+            "output_token_price_per_million": max(
+                float(v.get("output_token_price_per_million", 0.0))
+                for v in self.cost_config.values()
+            ),
+        }
 
     def has_price(self, model_name: str) -> bool:
         """Whether a price is known for this model name. Used to decide whether a
@@ -248,14 +271,22 @@ def calculate_inference_cost(
         cost_info = self._lookup_cost_info(model_name)
 
         if not cost_info:
-            print(
-                f"Warning: No cost configuration found for model {model_name} (lookup: {cost_lookup_name})"
-            )
-            if len(self.cost_config) > 0:
+            # Unmatched model: charge the highest price in the table rather than
+            # dropping the cost, and warn once per model name.
+            cost_info = self.max_price_info()
+            with self._unpriced_lock:
+                first_time = cost_lookup_name not in self.unpriced_models
+                self.unpriced_models[cost_lookup_name] = (
+                    self.unpriced_models.get(cost_lookup_name, 0) + 1
+                )
+            if first_time:
                 print(
-                    f"Available cost config keys (first 10): {list(self.cost_config.keys())[:10]}"
+                    f"Warning: No price profile for model {cost_lookup_name!r}; "
+                    "charging the maximum price "
+                    f"(${cost_info['input_token_price_per_million']}/"
+                    f"${cost_info['output_token_price_per_million']} per 1M tokens). "
+                    "Add it to model_cost/model_profiles.yaml."
                 )
-            return 0.0
 
         # Calculate cost
         input_tokens = token_usage.get("input_tokens", 0) or 0
diff --git a/llm_evaluation/run.py b/llm_evaluation/run.py
index fbbc3f1d..c72fca01 100644
--- a/llm_evaluation/run.py
+++ b/llm_evaluation/run.py
@@ -362,14 +362,18 @@ def evaluate_single_prediction(
         )
         return False
 
-    # Convert to universal model name
+    # Convert to universal model name. An unknown name is still graded (the score
+    # does not depend on it) and is charged the maximum price below, so a router
+    # cannot drop rows from its results by naming an unregistered model.
     try:
         universal_model_name = ModelNameManager.get_universal_name(model_name)
     except Exception as e:
-        logger.error(
-            f"Error converting model name '{model_name}' to universal name: {e}"
+        # Logged once per name by the evaluator and summarised after the run.
+        logger.debug(
+            f"Unknown model name '{model_name}' ({e}); evaluating it anyway "
+            "and charging the maximum price."
         )
-        return False
+        universal_model_name = model_name
 
     # Determine dataset name from global_index
     dataset_name = evaluator.determine_dataset_from_global_index(global_index)
@@ -586,6 +590,18 @@ def save_callback():
     logger.info(
         f"Predictions saved to: ./router_inference/predictions/{router_name}.json"
     )
+    if evaluator.unpriced_models:
+        max_price = evaluator.max_price_info()
+        logger.warning(
+            "Charged at the maximum price "
+            f"(${max_price['input_token_price_per_million']}/"
+            f"${max_price['output_token_price_per_million']} per 1M tokens) "
+            "because no price profile matched:"
+        )
+        for name, rows in sorted(
+            evaluator.unpriced_models.items(), key=lambda kv: -kv[1]
+        ):
+            logger.warning(f"  {name}: {rows} rows")
     logger.info("=" * 60)
 
     # Compute and display router-level metrics
diff --git a/model_cost/model_cost.json b/model_cost/model_cost.json
index ae82c1d1..debee82c 100644
--- a/model_cost/model_cost.json
+++ b/model_cost/model_cost.json
@@ -11,10 +11,18 @@
     "input_token_price_per_million": 0.5,
     "output_token_price_per_million": 3
   },
+  "google/gemini-3-flash-preview": {
+    "input_token_price_per_million": 0.5,
+    "output_token_price_per_million": 3
+  },
   "o4-mini": {
     "input_token_price_per_million": 1.1,
     "output_token_price_per_million": 4.4
   },
+  "o4-mini-2025-04-16": {
+    "input_token_price_per_million": 1.1,
+    "output_token_price_per_million": 4.4
+  },
   "gpt-5.1": {
     "input_token_price_per_million": 1.25,
     "output_token_price_per_million": 10.0
@@ -31,10 +39,18 @@
     "input_token_price_per_million": 0.25,
     "output_token_price_per_million": 2.0
   },
+  "openai/gpt-5-mini": {
+    "input_token_price_per_million": 0.25,
+    "output_token_price_per_million": 2.0
+  },
   "gpt-5-nano": {
     "input_token_price_per_million": 0.05,
     "output_token_price_per_million": 0.4
   },
+  "gpt-5-nano-2025-08-07": {
+    "input_token_price_per_million": 0.05,
+    "output_token_price_per_million": 0.4
+  },
   "deepseek-reasoner": {
     "input_token_price_per_million": 0.28,
     "output_token_price_per_million": 0.42
@@ -43,10 +59,6 @@
     "input_token_price_per_million": 2.0,
     "output_token_price_per_million": 12.0
   },
-  "deepseek-v3.2": {
-    "input_token_price_per_million": 0.28,
-    "output_token_price_per_million": 0.42
-  },
   "test": {
     "input_token_price_per_million": 1.0,
     "output_token_price_per_million": 2.0
@@ -55,6 +67,10 @@
     "input_token_price_per_million": 3.0,
     "output_token_price_per_million": 15.0
   },
+  "anthropic/claude-sonnet-4.5": {
+    "input_token_price_per_million": 3.0,
+    "output_token_price_per_million": 15.0
+  },
   "claude-3-haiku-20240307": {
     "input_token_price_per_million": 0.25,
     "output_token_price_per_million": 1.25
@@ -63,6 +79,10 @@
     "input_token_price_per_million": 0.15,
     "output_token_price_per_million": 0.6
   },
+  "openai/gpt-4o-mini": {
+    "input_token_price_per_million": 0.15,
+    "output_token_price_per_million": 0.6
+  },
   "glm-4.5-air": {
     "input_token_price_per_million": 0.1,
     "output_token_price_per_million": 0.3
@@ -75,14 +95,14 @@
     "input_token_price_per_million": 0.07,
     "output_token_price_per_million": 0.07
   },
+  "glm-4-air": {
+    "input_token_price_per_million": 0.07,
+    "output_token_price_per_million": 0.07
+  },
   "qwen_qwen3-vl-235b-a22b-thinking": {
     "input_token_price_per_million": 0.3,
     "output_token_price_per_million": 1.2
   },
-  "gemini-2.5-flash-lite": {
-    "input_token_price_per_million": 0.1,
-    "output_token_price_per_million": 0.4
-  },
   "gemini-2.5-flash": {
     "input_token_price_per_million": 0.3,
     "output_token_price_per_million": 2.5
@@ -95,10 +115,6 @@
     "input_token_price_per_million": 0.5,
     "output_token_price_per_million": 1.5
   },
-  "z-ai_glm-4.7": {
-    "input_token_price_per_million": 0.4,
-    "output_token_price_per_million": 1.5
-  },
   "qwen_qwen3-vl-235b-a22b-instruct": {
     "input_token_price_per_million": 0.2,
     "output_token_price_per_million": 1.2
@@ -115,22 +131,10 @@
     "input_token_price_per_million": 0.1,
     "output_token_price_per_million": 0.3
   },
-  "openai_gpt-oss-120b": {
-    "input_token_price_per_million": 0.039,
-    "output_token_price_per_million": 0.19
-  },
-  "qwen_qwen3-235b-a22b-2507": {
-    "input_token_price_per_million": 0.071,
-    "output_token_price_per_million": 0.463
-  },
   "qwen/qwen3-next-80b-a3b-instruct": {
     "input_token_price_per_million": 0.09,
     "output_token_price_per_million": 1.1
   },
-  "claude-haiku-4.5": {
-    "input_token_price_per_million": 1.0,
-    "output_token_price_per_million": 5.0
-  },
   "x-ai_grok-4.1-fast": {
     "input_token_price_per_million": 0.2,
     "output_token_price_per_million": 0.5
@@ -163,10 +167,6 @@
     "input_token_price_per_million": 0.2,
     "output_token_price_per_million": 0.2
   },
-  "gpt-4o": {
-    "input_token_price_per_million": 2.5,
-    "output_token_price_per_million": 10.0
-  },
   "qwen/qwen3-30b-a3b-instruct-2507": {
     "input_token_price_per_million": 0.08,
     "output_token_price_per_million": 0.33
@@ -203,6 +203,10 @@
     "input_token_price_per_million": 5.0,
     "output_token_price_per_million": 30.0
   },
+  "gpt-5.5-2026-04-23": {
+    "input_token_price_per_million": 5.0,
+    "output_token_price_per_million": 30.0
+  },
   "gpt-5.4-mini": {
     "input_token_price_per_million": 0.4,
     "output_token_price_per_million": 1.6
@@ -235,6 +239,10 @@
     "input_token_price_per_million": 0.071,
     "output_token_price_per_million": 0.1
   },
+  "qwen_qwen3-235b-a22b-2507": {
+    "input_token_price_per_million": 0.071,
+    "output_token_price_per_million": 0.1
+  },
   "qwen/qwen3.5-9b": {
     "input_token_price_per_million": 0.05,
     "output_token_price_per_million": 0.15
@@ -243,14 +251,26 @@
     "input_token_price_per_million": 0.26,
     "output_token_price_per_million": 0.38
   },
+  "deepseek-v3.2": {
+    "input_token_price_per_million": 0.26,
+    "output_token_price_per_million": 0.38
+  },
   "openai/gpt-4o": {
     "input_token_price_per_million": 2.5,
     "output_token_price_per_million": 10.0
   },
+  "gpt-4o": {
+    "input_token_price_per_million": 2.5,
+    "output_token_price_per_million": 10.0
+  },
   "claude-haiku-4-5-20251001": {
     "input_token_price_per_million": 1.0,
     "output_token_price_per_million": 5.0
   },
+  "claude-haiku-4.5": {
+    "input_token_price_per_million": 1.0,
+    "output_token_price_per_million": 5.0
+  },
   "claude-sonnet-4": {
     "input_token_price_per_million": 3.0,
     "output_token_price_per_million": 15.0
@@ -295,6 +315,14 @@
     "input_token_price_per_million": 0.15,
     "output_token_price_per_million": 0.6
   },
+  "openai_gpt-oss-120b": {
+    "input_token_price_per_million": 0.15,
+    "output_token_price_per_million": 0.6
+  },
+  "openai/gpt-oss-120b": {
+    "input_token_price_per_million": 0.15,
+    "output_token_price_per_million": 0.6
+  },
   "llama-4-maverick-17b-128e-instruct-fp8": {
     "input_token_price_per_million": 0.25,
     "output_token_price_per_million": 1.0
@@ -331,10 +359,18 @@
     "input_token_price_per_million": 0.4,
     "output_token_price_per_million": 1.5
   },
+  "z-ai_glm-4.7": {
+    "input_token_price_per_million": 0.4,
+    "output_token_price_per_million": 1.5
+  },
   "MiniMax-M3": {
     "input_token_price_per_million": 0.6,
     "output_token_price_per_million": 2.4
   },
+  "MiniMaxAI/MiniMax-M3": {
+    "input_token_price_per_million": 0.6,
+    "output_token_price_per_million": 2.4
+  },
   "agnes-2.0-flash": {
     "input_token_price_per_million": 0.03,
     "output_token_price_per_million": 0.15
@@ -347,11 +383,11 @@
     "input_token_price_per_million": 0.18,
     "output_token_price_per_million": 0.85
   },
-  "grok-4.3": {
+  "x-ai/grok-4.3": {
     "input_token_price_per_million": 1.25,
     "output_token_price_per_million": 2.5
   },
-  "x-ai/grok-4.3": {
+  "grok-4.3": {
     "input_token_price_per_million": 1.25,
     "output_token_price_per_million": 2.5
   },
@@ -359,26 +395,22 @@
     "input_token_price_per_million": 0.03,
     "output_token_price_per_million": 0.13
   },
+  "openai/gpt-oss-20b": {
+    "input_token_price_per_million": 0.03,
+    "output_token_price_per_million": 0.13
+  },
   "xiaomi/mimo-v2.5": {
     "input_token_price_per_million": 0.14,
     "output_token_price_per_million": 0.28
   },
-  "openai/gpt-5-mini": {
-    "input_token_price_per_million": 0.25,
-    "output_token_price_per_million": 2.0
-  },
-  "anthropic/claude-sonnet-4.5": {
-    "input_token_price_per_million": 3.0,
-    "output_token_price_per_million": 15.0
-  },
-  "openai/gpt-4o-mini": {
-    "input_token_price_per_million": 0.15,
-    "output_token_price_per_million": 0.6
-  },
   "google/gemini-2.5-flash-lite": {
     "input_token_price_per_million": 0.1,
     "output_token_price_per_million": 0.4
   },
+  "gemini-2.5-flash-lite": {
+    "input_token_price_per_million": 0.1,
+    "output_token_price_per_million": 0.4
+  },
   "google/gemini-2.5-pro": {
     "input_token_price_per_million": 1.25,
     "output_token_price_per_million": 10.0
diff --git a/model_cost/model_profiles.yaml b/model_cost/model_profiles.yaml
new file mode 100644
index 00000000..da9dce72
--- /dev/null
+++ b/model_cost/model_profiles.yaml
@@ -0,0 +1,389 @@
+# SPDX-FileCopyrightText: Copyright contributors to the RouterArena project
+# SPDX-License-Identifier: Apache-2.0
+#
+# Model price profiles: the source of truth for model_cost/model_cost.json.
+# One entry per model. After editing, regenerate the JSON with:
+#   python scripts/pricing/build_model_cost.py
+#
+# Fields
+#   price.input / price.output  USD per 1M tokens (required)
+#   aliases                     other spellings of the same model; they share its price
+#   as_of                       date the price was recorded
+#   source                      URL of the published price (required for new models)
+#   note                        free text, e.g. known conflicts to resolve
+#
+# Prices are fixed for now; a market sync will update them later. A model name that
+# matches no id or alias is charged the highest input and output price in this file.
+
+models:
+  "google/gemma-3n-e4b-it":
+    price: {input: 0.02, output: 0.04}
+    as_of: "2026-09-27"
+    source: null
+  "gemini-3-pro-preview":
+    price: {input: 2.0, output: 12.0}
+    as_of: "2026-09-27"
+    source: null
+  "gemini-3-flash-preview":
+    price: {input: 0.5, output: 3}
+    aliases: ["google/gemini-3-flash-preview"]
+    as_of: "2026-09-27"
+    source: null
+  "o4-mini":
+    price: {input: 1.1, output: 4.4}
+    aliases: ["o4-mini-2025-04-16"]
+    as_of: "2026-09-27"
+    source: null
+  "gpt-5.1":
+    price: {input: 1.25, output: 10.0}
+    as_of: "2026-09-27"
+    source: null
+  "gpt-5.2":
+    price: {input: 1.75, output: 14.0}
+    as_of: "2026-09-27"
+    source: null
+  "gpt-5.1-None":
+    price: {input: 1.25, output: 10.0}
+    as_of: "2026-09-27"
+    source: null
+  "gpt-5-mini":
+    price: {input: 0.25, output: 2.0}
+    aliases: ["openai/gpt-5-mini"]
+    as_of: "2026-09-27"
+    source: null
+  "gpt-5-nano":
+    price: {input: 0.05, output: 0.4}
+    aliases: ["gpt-5-nano-2025-08-07"]
+    as_of: "2026-09-27"
+    source: null
+  "deepseek-reasoner":
+    price: {input: 0.28, output: 0.42}
+    as_of: "2026-09-27"
+    source: null
+  "gemini-3.0-pro":
+    price: {input: 2.0, output: 12.0}
+    as_of: "2026-09-27"
+    source: null
+  "test":
+    price: {input: 1.0, output: 2.0}
+    as_of: "2026-09-27"
+    source: null
+  "claude-sonnet-4-5":
+    price: {input: 3.0, output: 15.0}
+    aliases: ["anthropic/claude-sonnet-4.5"]
+    as_of: "2026-09-27"
+    source: null
+  "claude-3-haiku-20240307":
+    price: {input: 0.25, output: 1.25}
+    as_of: "2026-09-27"
+    source: null
+  "gpt-4o-mini":
+    price: {input: 0.15, output: 0.6}
+    aliases: ["openai/gpt-4o-mini"]
+    as_of: "2026-09-27"
+    source: null
+  "glm-4.5-air":
+    price: {input: 0.1, output: 0.3}
+    as_of: "2026-09-27"
+    source: null
+  "glm-4.6":
+    price: {input: 0.28, output: 1.12}
+    as_of: "2026-09-27"
+    source: null
+  "glm-4-air-250414":
+    price: {input: 0.07, output: 0.07}
+    aliases: ["glm-4-air"]
+    as_of: "2026-09-27"
+    source: null
+    note: "Alias glm-4-air was previously matched by substring; kept at this price."
+  "qwen_qwen3-vl-235b-a22b-thinking":
+    price: {input: 0.3, output: 1.2}
+    as_of: "2026-09-27"
+    source: null
+  "gemini-2.5-flash":
+    price: {input: 0.3, output: 2.5}
+    as_of: "2026-09-27"
+    source: null
+  "gemini-2.0-flash-001":
+    price: {input: 0.1, output: 0.4}
+    as_of: "2026-09-27"
+    source: null
+  "qwen_qwen3-vl-32b-instruct":
+    price: {input: 0.5, output: 1.5}
+    as_of: "2026-09-27"
+    source: null
+  "qwen_qwen3-vl-235b-a22b-instruct":
+    price: {input: 0.2, output: 1.2}
+    as_of: "2026-09-27"
+    source: null
+  "qwen_qwen3-coder":
+    price: {input: 0.22, output: 0.95}
+    as_of: "2026-09-27"
+    source: null
+  "x-ai_grok-code-fast-1":
+    price: {input: 0.2, output: 1.5}
+    as_of: "2026-09-27"
+    source: null
+  "xiaomi_mimo-v2-flash:free":
+    price: {input: 0.1, output: 0.3}
+    as_of: "2026-09-27"
+    source: null
+  "qwen/qwen3-next-80b-a3b-instruct":
+    price: {input: 0.09, output: 1.1}
+    as_of: "2026-09-27"
+    source: null
+  "x-ai_grok-4.1-fast":
+    price: {input: 0.2, output: 0.5}
+    as_of: "2026-09-27"
+    source: null
+  "mistralai_devstral-2512:free":
+    price: {input: 0.05, output: 0.22}
+    as_of: "2026-09-27"
+    source: null
+  "meta-llama_llama-3.3-70b-instruct":
+    price: {input: 0.1, output: 0.32}
+    as_of: "2026-09-27"
+    source: null
+  "meta-llama_llama-3.1-405b-instruct":
+    price: {input: 3.5, output: 3.5}
+    as_of: "2026-09-27"
+    source: null
+  "mistralai/ministral-3b-2512":
+    price: {input: 0.1, output: 0.1}
+    as_of: "2026-09-27"
+    source: null
+  "mistralai/ministral-3-3b-2512":
+    price: {input: 0.1, output: 0.1}
+    as_of: "2026-09-27"
+    source: null
+  "mistralai/ministral-3-8b-2512":
+    price: {input: 0.15, output: 0.15}
+    as_of: "2026-09-27"
+    source: null
+  "mistralai/ministral-3-14b-2512":
+    price: {input: 0.2, output: 0.2}
+    as_of: "2026-09-27"
+    source: null
+  "qwen/qwen3-30b-a3b-instruct-2507":
+    price: {input: 0.08, output: 0.33}
+    as_of: "2026-09-27"
+    source: null
+    note: "Conflict: also profiled as qwen3-30b-a3b-instruct-2507 at 0.15/0.60. Unify at market sync."
+  "Qwen/Qwen3-Coder-Next":
+    price: {input: 0.07, output: 0.3}
+    as_of: "2026-09-27"
+    source: null
+  "qwen/qwen3-coder-30b-a3b-instruct":
+    price: {input: 0.07, output: 0.27}
+    as_of: "2026-09-27"
+    source: null
+  "moonshotai/kimi-k2.5":
+    price: {input: 0.6, output: 3.0}
+    as_of: "2026-09-27"
+    source: null
+  "z-ai/glm-5":
+    price: {input: 1.0, output: 3.2}
+    as_of: "2026-09-27"
+    source: null
+  "google/gemini-3.1-flash-lite":
+    price: {input: 0.25, output: 1.5}
+    as_of: "2026-09-27"
+    source: null
+  "claude-opus-4-7":
+    price: {input: 15.0, output: 75.0}
+    as_of: "2026-09-27"
+    source: null
+  "claude-haiku-4-5":
+    price: {input: 0.8, output: 4.0}
+    as_of: "2026-09-27"
+    source: null
+    note: "Conflict: same model as claude-haiku-4-5-20251001 (1.00/5.00). Kept separate so current scores do not move. Unify at market sync."
+  "gpt-5.5":
+    price: {input: 5.0, output: 30.0}
+    aliases: ["gpt-5.5-2026-04-23"]
+    as_of: "2026-09-27"
+    source: null
+  "gpt-5.4-mini":
+    price: {input: 0.4, output: 1.6}
+    as_of: "2026-09-27"
+    source: null
+  "gpt-4.1":
+    price: {input: 2.0, output: 8.0}
+    as_of: "2026-09-27"
+    source: null
+  "gemini-3.1-pro-preview":
+    price: {input: 2.0, output: 12.0}
+    as_of: "2026-09-27"
+    source: null
+  "gemini-3.1-flash-lite-preview":
+    price: {input: 0.1, output: 0.4}
+    as_of: "2026-09-27"
+    source: null
+  "deepseek/deepseek-v4-pro":
+    price: {input: 0.435, output: 0.87}
+    as_of: "2026-09-27"
+    source: null
+  "qwen/qwen3.5-flash-02-23":
+    price: {input: 0.065, output: 0.26}
+    as_of: "2026-09-27"
+    source: null
+  "deepseek/deepseek-v4-flash":
+    price: {input: 0.14, output: 0.28}
+    as_of: "2026-09-27"
+    source: null
+  "qwen/qwen3-235b-a22b-2507":
+    price: {input: 0.071, output: 0.1}
+    aliases: ["qwen_qwen3-235b-a22b-2507"]
+    as_of: "2026-09-27"
+    source: null
+    note: "Conflict: also profiled as qwen3-235b-a22b-instruct-2507 at 0.50/2.00 (charged to OrcaRouter). Dropped unused duplicate price 0.071/0.463 (qwen_qwen3-235b-a22b-2507). Unify at market sync."
+  "qwen/qwen3.5-9b":
+    price: {input: 0.05, output: 0.15}
+    as_of: "2026-09-27"
+    source: null
+  "deepseek/deepseek-v3.2":
+    price: {input: 0.26, output: 0.38}
+    aliases: ["deepseek-v3.2"]
+    as_of: "2026-09-27"
+    source: null
+    note: "Dropped unused duplicate price 0.28/0.42 (deepseek-v3.2)."
+  "openai/gpt-4o":
+    price: {input: 2.5, output: 10.0}
+    aliases: ["gpt-4o"]
+    as_of: "2026-09-27"
+    source: null
+  "claude-haiku-4-5-20251001":
+    price: {input: 1.0, output: 5.0}
+    aliases: ["claude-haiku-4.5"]
+    as_of: "2026-09-27"
+    source: null
+    note: "Conflict: also profiled as claude-haiku-4-5 at 0.80/4.00 (charged to Nadir Router). Unify at market sync."
+  "claude-sonnet-4":
+    price: {input: 3.0, output: 15.0}
+    as_of: "2026-09-27"
+    source: null
+  "deepseek-chat":
+    price: {input: 0.27, output: 1.1}
+    as_of: "2026-09-27"
+    source: null
+  "qwen3-235b-a22b-instruct-2507":
+    price: {input: 0.5, output: 2.0}
+    as_of: "2026-09-27"
+    source: null
+    note: "Conflict: same model as qwen/qwen3-235b-a22b-2507 (0.071/0.10). Kept separate so current scores do not move. Unify at market sync."
+  "qwen3-30b-a3b-instruct-2507":
+    price: {input: 0.15, output: 0.6}
+    as_of: "2026-09-27"
+    source: null
+    note: "Conflict: same model as qwen/qwen3-30b-a3b-instruct-2507 (0.08/0.33). Kept separate so current scores do not move. Unify at market sync."
+  "gpt-4.1-mini":
+    price: {input: 0.4, output: 1.6}
+    as_of: "2026-09-27"
+    source: null
+  "gpt-4.1-nano":
+    price: {input: 0.1, output: 0.4}
+    as_of: "2026-09-27"
+    source: null
+  "deepseek-v3.1":
+    price: {input: 1.23, output: 4.94}
+    as_of: "2026-09-27"
+    source: null
+  "gpt-5":
+    price: {input: 1.25, output: 10.0}
+    as_of: "2026-09-27"
+    source: null
+  "gpt-5-chat":
+    price: {input: 1.25, output: 10.0}
+    as_of: "2026-09-27"
+    source: null
+  "gpt-5.2-chat":
+    price: {input: 1.75, output: 14.0}
+    as_of: "2026-09-27"
+    source: null
+  "gpt-oss-120b":
+    price: {input: 0.15, output: 0.6}
+    aliases: ["openai_gpt-oss-120b", "openai/gpt-oss-120b"]
+    as_of: "2026-09-27"
+    source: null
+    note: "Dropped unused duplicate price 0.039/0.19 (openai_gpt-oss-120b)."
+  "llama-4-maverick-17b-128e-instruct-fp8":
+    price: {input: 0.25, output: 1.0}
+    as_of: "2026-09-27"
+    source: null
+  "grok-4":
+    price: {input: 5.5, output: 27.5}
+    as_of: "2026-09-27"
+    source: null
+  "claude-opus-4-1":
+    price: {input: 15.0, output: 75.0}
+    as_of: "2026-09-27"
+    source: null
+  "claude-opus-4-6":
+    price: {input: 5.0, output: 25.0}
+    as_of: "2026-09-27"
+    source: null
+  "gpt-5.4-nano":
+    price: {input: 0.2, output: 1.25}
+    as_of: "2026-09-27"
+    source: null
+  "grok-4-1-fast-reasoning":
+    price: {input: 0.2, output: 0.5}
+    as_of: "2026-09-27"
+    source: null
+  "gpt-5.3-chat":
+    price: {input: 1.75, output: 14.0}
+    as_of: "2026-09-27"
+    source: null
+  "gpt-5.4":
+    price: {input: 2.5, output: 15.0}
+    as_of: "2026-09-27"
+    source: null
+  "z-ai/glm-4.7":
+    price: {input: 0.4, output: 1.5}
+    aliases: ["z-ai_glm-4.7"]
+    as_of: "2026-09-27"
+    source: null
+  "MiniMax-M3":
+    price: {input: 0.6, output: 2.4}
+    aliases: ["MiniMaxAI/MiniMax-M3"]
+    as_of: "2026-09-27"
+    source: null
+  "agnes-2.0-flash":
+    price: {input: 0.03, output: 0.15}
+    as_of: "2026-09-27"
+    source: null
+  "THUDM/GLM-4-9B-0414":
+    price: {input: 0.05, output: 0.61}
+    as_of: "2026-09-27"
+    source: null
+  "deepseek-ai/DeepSeek-R1-0528-Qwen3-8B":
+    price: {input: 0.18, output: 0.85}
+    as_of: "2026-09-27"
+    source: null
+  "x-ai/grok-4.3":
+    price: {input: 1.25, output: 2.5}
+    aliases: ["grok-4.3"]
+    as_of: "2026-09-27"
+    source: null
+  "gpt-oss-20b":
+    price: {input: 0.03, output: 0.13}
+    aliases: ["openai/gpt-oss-20b"]
+    as_of: "2026-09-27"
+    source: null
+  "xiaomi/mimo-v2.5":
+    price: {input: 0.14, output: 0.28}
+    as_of: "2026-09-27"
+    source: null
+  "google/gemini-2.5-flash-lite":
+    price: {input: 0.1, output: 0.4}
+    aliases: ["gemini-2.5-flash-lite"]
+    as_of: "2026-09-27"
+    source: null
+  "google/gemini-2.5-pro":
+    price: {input: 1.25, output: 10.0}
+    as_of: "2026-09-27"
+    source: null
+  "google/gemma-4-31b-it":
+    price: {input: 0.08, output: 0.35}
+    as_of: "2026-09-27"
+    source: null
diff --git a/router_inference/check_config_prediction_files.py b/router_inference/check_config_prediction_files.py
index f4d09823..c1c7857d 100644
--- a/router_inference/check_config_prediction_files.py
+++ b/router_inference/check_config_prediction_files.py
@@ -176,49 +176,46 @@ def check_model_costs(
     for model in config_models:
         used_models.add(model)
 
-    # Check each model has cost configuration
+    # Check each model has a price profile (exact id or alias match only; see
+    # model_cost/model_profiles.yaml). A model without one is not rejected: the
+    # evaluator charges it the highest price in the table, so we warn loudly.
     missing_costs = []
-    for model_name in used_models:
+    for model_name in sorted(used_models):
         try:
             # Convert to universal name for cost lookup
             universal_name = model_manager.get_universal_name(model_name)
+        except Exception:
+            universal_name = model_name
 
-            # Remove _batch suffix if present
-            cost_lookup_name = universal_name
-            if universal_name.endswith("_batch"):
-                cost_lookup_name = universal_name[:-6]
+        # Remove _batch suffix if present
+        cost_lookup_name = universal_name
+        if universal_name.endswith("_batch"):
+            cost_lookup_name = universal_name[:-6]
 
-            # Check if cost exists (exact match or partial match)
-            has_cost = False
-            if cost_lookup_name in cost_config:
-                has_cost = True
-            else:
-                # Try partial matches
-                for config_name in cost_config.keys():
-                    if (
-                        config_name in cost_lookup_name
-                        or cost_lookup_name in config_name
-                    ):
-                        has_cost = True
-                        break
-
-            if not has_cost:
-                missing_costs.append(f"{model_name} (universal: {cost_lookup_name})")
-        except Exception as e:
-            # If we can't convert, try original name
-            if model_name not in cost_config:
-                missing_costs.append(f"{model_name} (conversion failed: {str(e)})")
+        if cost_lookup_name not in cost_config:
+            missing_costs.append(f"{model_name} (looked up as: {cost_lookup_name})")
 
     if missing_costs:
-        errors.append(f"Missing cost configuration for {len(missing_costs)} model(s):")
+        max_in = max(
+            float(v.get("input_token_price_per_million", 0.0))
+            for v in cost_config.values()
+        )
+        max_out = max(
+            float(v.get("output_token_price_per_million", 0.0))
+            for v in cost_config.values()
+        )
+        errors.append(f"No price profile for {len(missing_costs)} model(s):")
         for model in missing_costs:
             errors.append(f"  - {model}")
         errors.append(
-            "\nPlease add cost configuration to model_cost/model_cost.json or update "
-            "your router config to use models with existing cost configurations."
+            f"These rows will be charged the maximum price (${max_in}/${max_out} per "
+            "1M tokens). To avoid that, add a profile (with a source link) to "
+            "model_cost/model_profiles.yaml and run "
+            "`python scripts/pricing/build_model_cost.py`."
         )
 
-    return len(missing_costs) == 0, errors
+    # Missing prices are a warning, not a failure.
+    return True, errors
 
 
 # Model slugs that upstream providers have retired and silently redirect to a
@@ -659,7 +656,10 @@ def main():
     try:
         if predictions is not None and config is not None:
             cost_valid, cost_errors = check_model_costs(predictions, config)
-            if cost_valid:
+            if cost_valid and cost_errors:
+                for warning in cost_errors:
+                    print(f"  ⚠ {warning}")
+            elif cost_valid:
                 cost_config = load_cost_config()
                 print(
                     f"✓ All models have cost configurations ({len(cost_config)} models in cost file)"
diff --git a/scripts/pricing/build_model_cost.py b/scripts/pricing/build_model_cost.py
new file mode 100644
index 00000000..dcbd3c50
--- /dev/null
+++ b/scripts/pricing/build_model_cost.py
@@ -0,0 +1,124 @@
+# SPDX-FileCopyrightText: Copyright contributors to the RouterArena project
+# SPDX-License-Identifier: Apache-2.0
+
+"""Generate ``model_cost/model_cost.json`` from ``model_cost/model_profiles.yaml``.
+
+``model_profiles.yaml`` is the source of truth for model prices: one profile per
+model, with optional aliases (other spellings of the same model). The evaluator
+and the submission checks still read ``model_cost.json``, so this script expands
+every profile into one JSON entry for its id and one for each alias, all at the
+profile's price.
+
+Usage:
+    python scripts/pricing/build_model_cost.py          # rewrite model_cost.json
+    python scripts/pricing/build_model_cost.py --check  # exit 1 if it is out of date
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import sys
+from pathlib import Path
+from typing import Any, Dict, List, Optional
+
+import yaml
+
+REPO_ROOT = Path(__file__).resolve().parents[2]
+PROFILES_PATH = REPO_ROOT / "model_cost" / "model_profiles.yaml"
+COST_PATH = REPO_ROOT / "model_cost" / "model_cost.json"
+
+
+def load_profiles(path: Path = PROFILES_PATH) -> Dict[str, Dict[str, Any]]:
+    """Load and validate the profile file. Raises ValueError on a bad entry."""
+    with open(path, "r", encoding="utf-8") as f:
+        data = yaml.safe_load(f) or {}
+    models = data.get("models")
+    if not isinstance(models, dict) or not models:
+        raise ValueError(f"{path}: expected a non-empty 'models' mapping")
+
+    seen: Dict[str, str] = {}
+    errors: List[str] = []
+    for model_id, profile in models.items():
+        if not isinstance(profile, dict):
+            errors.append(f"{model_id}: profile must be a mapping")
+            continue
+        price = profile.get("price") or {}
+        for side in ("input", "output"):
+            value = price.get(side)
+            if (
+                not isinstance(value, (int, float))
+                or isinstance(value, bool)
+                or value < 0
+            ):
+                errors.append(f"{model_id}: price.{side} must be a number >= 0")
+        aliases = profile.get("aliases") or []
+        if not isinstance(aliases, list) or not all(
+            isinstance(a, str) for a in aliases
+        ):
+            errors.append(f"{model_id}: aliases must be a list of strings")
+            aliases = []
+        for name in [model_id, *aliases]:
+            if name in seen:
+                errors.append(
+                    f"{name!r} is used by both {seen[name]!r} and {model_id!r}"
+                )
+            seen[name] = model_id
+    if errors:
+        raise ValueError(f"{path}: invalid profiles:\n  " + "\n  ".join(errors))
+    return models
+
+
+def build_cost_table(models: Dict[str, Dict[str, Any]]) -> Dict[str, Dict[str, Any]]:
+    """Expand profiles into the flat ``model_cost.json`` mapping."""
+    table: Dict[str, Dict[str, Any]] = {}
+    for model_id, profile in models.items():
+        entry = {
+            "input_token_price_per_million": profile["price"]["input"],
+            "output_token_price_per_million": profile["price"]["output"],
+        }
+        for name in [model_id, *(profile.get("aliases") or [])]:
+            table[name] = dict(entry)
+    return table
+
+
+def render(table: Dict[str, Dict[str, Any]]) -> str:
+    return json.dumps(table, indent=2, ensure_ascii=False) + "\n"
+
+
+def main(argv: Optional[List[str]] = None) -> int:
+    parser = argparse.ArgumentParser(description=__doc__.splitlines()[0])
+    parser.add_argument(
+        "--check",
+        action="store_true",
+        help="Do not write; exit 1 if model_cost.json differs from the profiles.",
+    )
+    args = parser.parse_args(argv)
+
+    try:
+        expected = render(build_cost_table(load_profiles()))
+    except ValueError as e:
+        print(e, file=sys.stderr)
+        return 1
+
+    current = COST_PATH.read_text(encoding="utf-8") if COST_PATH.exists() else ""
+    if args.check:
+        if current != expected:
+            print(
+                "model_cost/model_cost.json is out of date with model_profiles.yaml. "
+                "Run: python scripts/pricing/build_model_cost.py",
+                file=sys.stderr,
+            )
+            return 1
+        print("model_cost.json is up to date.")
+        return 0
+
+    COST_PATH.write_text(expected, encoding="utf-8")
+    print(
+        f"Wrote {COST_PATH.relative_to(REPO_ROOT)} ({len(json.loads(expected))} entries)."
+    )
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/tests/test_pricing.py b/tests/test_pricing.py
new file mode 100644
index 00000000..f6ceb848
--- /dev/null
+++ b/tests/test_pricing.py
@@ -0,0 +1,93 @@
+# SPDX-FileCopyrightText: Copyright contributors to the RouterArena project
+# SPDX-License-Identifier: Apache-2.0
+
+"""Tests for model price profiles and the evaluator's price lookup."""
+
+import json
+import sys
+from pathlib import Path
+
+import pytest
+
+REPO_ROOT = Path(__file__).resolve().parents[1]
+sys.path.insert(0, str(REPO_ROOT))
+sys.path.insert(0, str(REPO_ROOT / "llm_evaluation"))
+sys.path.insert(0, str(REPO_ROOT / "scripts" / "pricing"))
+
+import build_model_cost  # noqa: E402
+from evaluate_models import ModelEvaluator  # noqa: E402
+
+USAGE = {
+    "input_tokens": 1_000_000,
+    "output_tokens": 1_000_000,
+    "total_tokens": 2_000_000,
+}
+
+
+@pytest.fixture(scope="module")
+def evaluator():
+    return ModelEvaluator(cached_results_dir=str(REPO_ROOT / "cached_results"))
+
+
+def test_model_cost_json_is_generated_from_profiles():
+    profiles = build_model_cost.load_profiles()
+    expected = build_model_cost.render(build_model_cost.build_cost_table(profiles))
+    actual = (REPO_ROOT / "model_cost" / "model_cost.json").read_text(encoding="utf-8")
+    assert actual == expected, "run: python scripts/pricing/build_model_cost.py"
+
+
+def test_every_name_is_unique_across_ids_and_aliases(tmp_path):
+    bad = tmp_path / "profiles.yaml"
+    bad.write_text(
+        "models:\n"
+        "  a: {price: {input: 1, output: 2}, aliases: [b]}\n"
+        "  b: {price: {input: 1, output: 2}}\n"
+    )
+    with pytest.raises(ValueError, match="used by both"):
+        build_model_cost.load_profiles(bad)
+
+
+def test_negative_or_missing_price_is_rejected(tmp_path):
+    bad = tmp_path / "profiles.yaml"
+    bad.write_text("models:\n  a: {price: {input: -1}}\n")
+    with pytest.raises(ValueError, match="price.input"):
+        build_model_cost.load_profiles(bad)
+
+
+def test_alias_is_charged_the_profile_price(evaluator):
+    # Lynkr names this model "openai/gpt-oss-120b"; its profile id is "gpt-oss-120b".
+    assert evaluator.has_price("openai/gpt-oss-120b")
+    assert evaluator.calculate_inference_cost(
+        "openai/gpt-oss-120b", USAGE
+    ) == pytest.approx(evaluator.calculate_inference_cost("gpt-oss-120b", USAGE))
+
+
+def test_no_substring_matching(evaluator):
+    # Used to match "gpt-oss-120b" because that key is a substring of the name.
+    assert not evaluator.has_price("gpt-oss-120b-some-finetune")
+    assert not evaluator.has_price("gpt")
+
+
+def test_unmatched_model_is_charged_the_maximum_price(evaluator):
+    max_price = evaluator.max_price_info()
+    table = json.loads((REPO_ROOT / "model_cost" / "model_cost.json").read_text())
+    assert max_price["input_token_price_per_million"] == max(
+        v["input_token_price_per_million"] for v in table.values()
+    )
+    assert max_price["output_token_price_per_million"] == max(
+        v["output_token_price_per_million"] for v in table.values()
+    )
+    cost = evaluator.calculate_inference_cost("not-a-registered-model", USAGE)
+    assert cost == pytest.approx(
+        max_price["input_token_price_per_million"]
+        + max_price["output_token_price_per_million"]
+    )
+    assert evaluator.unpriced_models.get("not-a-registered-model", 0) >= 1
+
+
+def test_unmatched_model_is_never_cheaper_than_a_registered_one(evaluator):
+    table = json.loads((REPO_ROOT / "model_cost" / "model_cost.json").read_text())
+    unmatched = evaluator.calculate_inference_cost("not-a-registered-model", USAGE)
+    assert all(
+        evaluator.calculate_inference_cost(name, USAGE) <= unmatched for name in table
+    )