diff --git a/litellm/config.yaml b/litellm/config.yaml
index 9cbb146..2c235e4 100644
--- a/litellm/config.yaml
+++ b/litellm/config.yaml
@@ -1,18 +1,104 @@
 model_list:
-  # --- OpenRouter models ---
+  # ═══════════════════════════════════════════════
+  # TIER 1: Free providers (try first)
+  # ═══════════════════════════════════════════════
+
+  # --- Groq (free tier, very fast) ---
+  - model_name: llama-3.3-70b
+    litellm_params:
+      model: groq/llama-3.3-70b-versatile
+      api_key: os.environ/GROQ_API_KEY
+
+  # --- Cerebras (free tier, very fast) ---
+  - model_name: llama-3.3-70b-cerebras
+    litellm_params:
+      model: cerebras/llama-3.3-70b
+      api_key: os.environ/CEREBRAS_API_KEY
+
+  # --- OpenRouter free models ---
+  - model_name: deepseek-v3-free
+    litellm_params:
+      model: openrouter/deepseek/deepseek-chat-v3-0324:free
+      api_key: os.environ/OPENROUTER_API_KEY
+
+  # ═══════════════════════════════════════════════
+  # TIER 2: SiliconFlow (cheapest paid, ~3-5x cheaper than OpenRouter)
+  # ═══════════════════════════════════════════════
+
+  # DeepSeek V3 — best value daily driver ($0.13 in / $0.28 out per M)
+  - model_name: deepseek-v3
+    litellm_params:
+      model: openai/deepseek-ai/DeepSeek-V3-0324
+      api_base: https://api.siliconflow.com/v1
+      api_key: os.environ/SILICONFLOW_API_KEY
+
+  # DeepSeek V3.2 via SiliconFlow (cheaper than DeepInfra)
+  - model_name: deepseek-v3.2
+    litellm_params:
+      model: openai/deepseek-ai/DeepSeek-V3-0324
+      api_base: https://api.siliconflow.com/v1
+      api_key: os.environ/SILICONFLOW_API_KEY
+
+  # GLM-4.7 via SiliconFlow
+  - model_name: glm-4.7
+    litellm_params:
+      model: openai/THUDM/GLM-4-32B-0414
+      api_base: https://api.siliconflow.com/v1
+      api_key: os.environ/SILICONFLOW_API_KEY
+
+  # Qwen3 Coder 480B MoE via SiliconFlow ($1.14 in / $2.28 out per M)
+  - model_name: qwen3-coder
+    litellm_params:
+      model: openai/Qwen/Qwen3-Coder-480B-A35B-Instruct
+      api_base: https://api.siliconflow.com/v1
+      api_key: os.environ/SILICONFLOW_API_KEY
+
+  # Qwen3 Coder 30B — cheaper alternative for simpler tasks
+  - model_name: qwen3-coder-30b
+    litellm_params:
+      model: openai/Qwen/Qwen3-Coder-30B-A3B-Instruct
+      api_base: https://api.siliconflow.com/v1
+      api_key: os.environ/SILICONFLOW_API_KEY
+
+  # ═══════════════════════════════════════════════
+  # TIER 3: DeepInfra (good mid-range pricing)
+  # ═══════════════════════════════════════════════
+
+  # DeepSeek V3 fallback (if SiliconFlow is down)
+  - model_name: deepseek-v3
+    litellm_params:
+      model: deepinfra/deepseek-ai/DeepSeek-V3-0324
+      api_key: os.environ/DEEPINFRA_API_KEY
+
+  - model_name: deepseek-r1
+    litellm_params:
+      model: deepinfra/deepseek-ai/DeepSeek-R1
+      api_key: os.environ/DEEPINFRA_API_KEY
+
+  - model_name: devstral
+    litellm_params:
+      model: deepinfra/mistralai/Devstral-Small-2505
+      api_key: os.environ/DEEPINFRA_API_KEY
+
+  # ═══════════════════════════════════════════════
+  # TIER 4: OpenRouter (most expensive, widest selection)
+  # ═══════════════════════════════════════════════
+
+  # Kimi K2.5 — DeepInfra is cheapest ($0.45 in / $2.25 out per M)
+  - model_name: kimi-k2.5
+    litellm_params:
+      model: deepinfra/moonshotai/Kimi-K2.5
+      api_key: os.environ/DEEPINFRA_API_KEY
+
+  # Kimi K2.5 fallback via OpenRouter
   - model_name: kimi-k2.5
     litellm_params:
       model: openrouter/moonshotai/kimi-k2.5
       api_key: os.environ/OPENROUTER_API_KEY
 
-  - model_name: devstral
+  - model_name: minimax-m2.5
     litellm_params:
-      model: openrouter/mistralai/devstral-small
-      api_key: os.environ/OPENROUTER_API_KEY
-
-  - model_name: minimax-m2
-    litellm_params:
-      model: openrouter/minimax/minimax-m1
+      model: openrouter/minimax/minimax-m2.5
       api_key: os.environ/OPENROUTER_API_KEY
 
   - model_name: gpt-oss
@@ -20,7 +106,17 @@ model_list:
       model: openrouter/openai/gpt-4.1-mini
       api_key: os.environ/OPENROUTER_API_KEY
 
-  # --- OpenRouter models (writing) ---
+  - model_name: gemini-3-flash-preview
+    litellm_params:
+      model: openrouter/google/gemini-3-flash-preview
+      api_key: os.environ/OPENROUTER_API_KEY
+
+  - model_name: trinity-large-preview
+    litellm_params:
+      model: openrouter/arcee-ai/trinity-large-preview
+      api_key: os.environ/OPENROUTER_API_KEY
+
+  # --- OpenRouter premium models ---
   - model_name: gemini-2.5-pro
     litellm_params:
       model: openrouter/google/gemini-2.5-pro-preview
@@ -36,66 +132,29 @@ model_list:
       model: openrouter/openai/gpt-4.1
       api_key: os.environ/OPENROUTER_API_KEY
 
+  # DeepSeek V3 last-resort fallback via OpenRouter
   - model_name: deepseek-v3
     litellm_params:
-      model: openrouter/deepseek/deepseek-chat-v3-0324:free
+      model: openrouter/deepseek/deepseek-chat-v3-0324
       api_key: os.environ/OPENROUTER_API_KEY
 
-  # --- SiliconFlow models ---
-  - model_name: glm-4.7
-    litellm_params:
-      model: openai/THUDM/GLM-4-32B-0414
-      api_base: https://api.siliconflow.com/v1
-      api_key: os.environ/SILICONFLOW_API_KEY
-
-  - model_name: qwen3-coder
-    litellm_params:
-      model: openrouter/qwen/qwen3-coder
-      api_key: os.environ/OPENROUTER_API_KEY
-
-  - model_name: qwen3-coder-480b-sf
-    litellm_params:
-      model: openai/Qwen/Qwen3-Coder-480B-A35B-Instruct
-      api_base: https://api.siliconflow.com/v1
-      api_key: os.environ/SILICONFLOW_API_KEY
-
-  - model_name: qwen3-coder-30b-sf
-    litellm_params:
-      model: openai/Qwen/Qwen3-Coder-30B-A3B-Instruct
-      api_base: https://api.siliconflow.com/v1
-      api_key: os.environ/SILICONFLOW_API_KEY
-
-  # --- DeepInfra models ---
-  - model_name: deepseek-v3.2
-    litellm_params:
-      model: deepinfra/deepseek-ai/DeepSeek-V3-0324
-      api_key: os.environ/DEEPINFRA_API_KEY
-
-  - model_name: devstral-deepinfra
-    litellm_params:
-      model: deepinfra/mistralai/Devstral-Small-2505
-      api_key: os.environ/DEEPINFRA_API_KEY
-
-  - model_name: deepseek-r1
-    litellm_params:
-      model: deepinfra/deepseek-ai/DeepSeek-R1
-      api_key: os.environ/DEEPINFRA_API_KEY
-
-  # --- Groq (free/fast) ---
-  - model_name: llama-3.3-70b
-    litellm_params:
-      model: groq/llama-3.3-70b-versatile
-      api_key: os.environ/GROQ_API_KEY
-
-  # --- Cerebras (free/fast) ---
-  - model_name: llama-3.3-70b-cerebras
-    litellm_params:
-      model: cerebras/llama-3.3-70b
-      api_key: os.environ/CEREBRAS_API_KEY
-
 general_settings:
   master_key: os.environ/LITELLM_MASTER_KEY
 
 litellm_settings:
   drop_params: true
   set_verbose: false
+  num_retries: 2
+  request_timeout: 600
+
+  # ── Response caching via Valkey (reuses SearXNG's instance) ──
+  cache: true
+  cache_params:
+    type: redis
+    host: valkey
+    port: 6379
+    ttl: 3600
+
+  # ── Budget limit: $3/day to prevent surprise bills ──
+  max_budget: 3.0
+  budget_duration: "1d"