Keep only 5 best Silicon Flow models for free tier

api/routes.py

@@ -45,25 +45,17 @@ REQUESTED_PROVIDER_MODELS = [
     # qwen-3-235b-a22b-instruct-2507 exists but is rate-limited
     # zai-glm-4.7 and gpt-oss-120b are not accessible with current key
     "cerebras/llama3.1-8b",
-    # Silicon Flow models (high rate limits, free tier)
-    # Qwen3.6 series - high speed multimodal
+    # Silicon Flow models (top 5 for free tier)
+    # DeepSeek-V4-Flash - best overall coding, 1M context, MoE
+    "silicon/deepseek-ai/DeepSeek-V4-Flash",
+    # Qwen3-Coder-480B - repository-scale coding, 256K+ context
+    "silicon/Qwen/Qwen3-Coder-480B-A35B-Instruct",
+    # Qwen3.6-35B-A3B - multimodal, 262K context
     "silicon/Qwen/Qwen3.6-35B-A3B",
-    "silicon/Qwen/Qwen3.6-27B",
-    # Qwen3.5 series - high throughput
-    "silicon/Qwen/Qwen3.5-35B-A3B",
-    "silicon/Qwen/Qwen3.5-27B",
-    "silicon/Qwen/Qwen2.5-7B-Instruct",
+    # Qwen2.5-72B - strong general purpose, 128K context
     "silicon/Qwen/Qwen2.5-72B-Instruct",
-    # Qwen3 Coder - best for coding
-    "silicon/Qwen/Qwen3-Coder-32B-A3B-Instruct",
-    "silicon/Qwen/Qwen3-Coder-480B-A35B-Instruct",
-    # DeepSeek - extremely high throughput MoE
-    "silicon/deepseek-ai/DeepSeek-V4-Flash",
-    "silicon/deepseek-ai/DeepSeek-V3",
-    # Google Gemma - fast and capable
-    "silicon/google/gemma-4-26B-A4B-it",
-    "silicon/google/gemma-4-31B-it",
-    "silicon/google/gemma-3-27B-it",
+    # QwQ-32B - best reasoning model
+    "silicon/Qwen/QwQ-32B",
     # Groq models (ultra fast inference)
     "groq/llama-3.3-70b-versatile",
     "groq/llama-3.1-8b-instant",