feat(gateway): update llm-gateway.yaml

llm-gateway.yaml

@@ -58,7 +58,91 @@ providers:
     timeout: 120s
 
 models:
-  # ── DeepSeek V3.2 ──
+  # ═══ TIER 1: Free (OpenRouter free models, $0) ═══
+  - name: "deepseek-v3-free"
+    routes:
+      - provider: openrouter
+        model: "deepseek/deepseek-v3.2-20251201:free"
+
+  - name: "gemini-2.5-flash-free"
+    routes:
+      - provider: openrouter
+        model: "google/gemini-2.5-flash:free"
+
+  - name: "gemini-2.0-flash-free"
+    routes:
+      - provider: openrouter
+        model: "google/gemini-2.0-flash-001:free"
+
+  - name: "gpt-oss-free"
+    routes:
+      - provider: openrouter
+        model: "openai/gpt-oss-120b:free"
+
+  - name: "gpt-5-nano-free"
+    routes:
+      - provider: openrouter
+        model: "openai/gpt-5-nano-2025-08-07:free"
+
+  - name: "grok-4.1-fast-free"
+    routes:
+      - provider: openrouter
+        model: "x-ai/grok-4.1-fast:free"
+
+  - name: "trinity-large-free"
+    routes:
+      - provider: openrouter
+        model: "arcee-ai/trinity-large-preview:free"
+
+  - name: "kimi-k2.5-free"
+    routes:
+      - provider: openrouter
+        model: "moonshotai/kimi-k2.5-0127:free"
+
+  # ═══ TIER 2: Low cost (Groq, Cerebras — free tier with rate limits) ═══
+  - name: "llama-3.3-70b"
+    routes:
+      - provider: groq
+        model: "llama-3.3-70b-versatile"
+        pricing: { input: 0.59, output: 0.79 }
+      - provider: deepinfra
+        model: "meta-llama/Llama-3.3-70B-Instruct"
+        pricing: { input: 0.23, output: 0.40 }
+
+  - name: "llama-3.1-8b"
+    routes:
+      - provider: groq
+        model: "llama-3.1-8b-instant"
+        pricing: { input: 0.05, output: 0.08 }
+      - provider: cerebras
+        model: "llama3.1-8b"
+        pricing: { input: 0.10, output: 0.10 }
+      - provider: deepinfra
+        model: "meta-llama/Meta-Llama-3.1-8B-Instruct"
+        pricing: { input: 0.03, output: 0.05 }
+
+  - name: "gpt-oss"
+    routes:
+      - provider: groq
+        model: "openai/gpt-oss-120b"
+        pricing: { input: 0.15, output: 0.60 }
+      - provider: cerebras
+        model: "gpt-oss-120b"
+        pricing: { input: 0.35, output: 0.75 }
+      - provider: deepinfra
+        model: "openai/gpt-oss-120b"
+        pricing: { input: 0.05, output: 0.24 }
+
+  - name: "gpt-oss-20b"
+    routes:
+      - provider: groq
+        model: "openai/gpt-oss-20b"
+        pricing: { input: 0.075, output: 0.30 }
+      - provider: deepinfra
+        model: "openai/gpt-oss-20b"
+        pricing: { input: 0.04, output: 0.16 }
+
+  # ═══ TIER 3: DeepSeek V3.2 (cheapest flagship) ═══
   - name: "deepseek-v3.2"
     routes:
       - provider: deepinfra
@@ -71,30 +155,7 @@ models:
         model: "deepseek/deepseek-chat-v3-0324"
         pricing: { input: 0.30, output: 0.88 }
 
-  # ── DeepSeek R1 ──
+  # ═══ TIER 4: Ultra-cheap DeepInfra ═══
-  - name: "deepseek-r1"
-    routes:
-      - provider: deepinfra
-        model: "deepseek-ai/DeepSeek-R1"
-        pricing: { input: 0.40, output: 1.60 }
-      - provider: openrouter
-        model: "deepseek/deepseek-r1"
-        pricing: { input: 0.55, output: 2.19 }
-
-  # ── GPT-OSS (OpenAI open-weight MoE) ──
-  - name: "gpt-oss"
-    routes:
-      - provider: deepinfra
-        model: "openai/gpt-oss-120b"
-        pricing: { input: 0.05, output: 0.24 }
-
-  - name: "gpt-oss-20b"
-    routes:
-      - provider: deepinfra
-        model: "openai/gpt-oss-20b"
-        pricing: { input: 0.04, output: 0.16 }
-
-  # ── Nemotron ──
   - name: "nemotron-super"
     routes:
       - provider: deepinfra
@@ -107,13 +168,28 @@ models:
         model: "nvidia/NVIDIA-Nemotron-Nano-9B-v2"
         pricing: { input: 0.04, output: 0.16 }
 
-  # ── Devstral ──
+  # ═══ TIER 5: DeepSeek R1 & reasoning ═══
+  - name: "deepseek-r1"
+    routes:
+      - provider: deepinfra
+        model: "deepseek-ai/DeepSeek-R1"
+        pricing: { input: 0.40, output: 1.60 }
+      - provider: openrouter
+        model: "deepseek/deepseek-r1"
+        pricing: { input: 0.55, output: 2.19 }
+
+  - name: "deepseek-r1-distill-qwen-32b"
+    routes:
+      - provider: deepinfra
+        model: "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B"
+        pricing: { input: 0.07, output: 0.16 }
+
   - name: "devstral"
     routes:
       - provider: deepinfra
         model: "mistralai/Devstral-Small-2505"
 
-  # ── GLM ──
+  # ═══ TIER 6: GLM ═══
   - name: "glm-4.6"
     routes:
       - provider: deepinfra
@@ -125,6 +201,9 @@ models:
       - provider: deepinfra
         model: "zai-org/GLM-4.7"
         pricing: { input: 0.40, output: 1.75 }
+      - provider: cerebras
+        model: "zai-glm-4.7"
+        pricing: { input: 2.25, output: 2.75 }
       - provider: siliconflow
         model: "THUDM/GLM-4-32B-0414"
 
@@ -134,7 +213,7 @@ models:
         model: "zai-org/GLM-5"
         pricing: { input: 0.80, output: 2.56 }
 
-  # ── Kimi ──
+  # ═══ TIER 7: Kimi ═══
   - name: "kimi-k2"
     routes:
       - provider: deepinfra
@@ -152,7 +231,7 @@ models:
       - provider: openrouter
         model: "moonshotai/kimi-k2.5"
 
-  # ── Qwen3 Coder ──
+  # ═══ TIER 8: SiliconFlow (Qwen) ═══
   - name: "qwen3-coder"
     routes:
       - provider: siliconflow
@@ -164,70 +243,7 @@ models:
       - provider: siliconflow
         model: "Qwen/Qwen3-Coder-30B-A3B-Instruct"
 
-  # ── Llama ──
+  # ═══ TIER 9: OpenRouter premium (paid) ═══
-  - name: "llama-3.3-70b"
-    routes:
-      - provider: groq
-        model: "llama-3.3-70b-versatile"
-      - provider: cerebras
-        model: "llama-3.3-70b"
-      - provider: deepinfra
-        model: "meta-llama/Llama-3.3-70B-Instruct"
-        pricing: { input: 0.23, output: 0.40 }
-
-  - name: "llama-3.1-8b"
-    routes:
-      - provider: groq
-        model: "llama-3.1-8b-instant"
-      - provider: cerebras
-        model: "llama-3.1-8b"
-      - provider: deepinfra
-        model: "meta-llama/Meta-Llama-3.1-8B-Instruct"
-        pricing: { input: 0.03, output: 0.05 }
-
-  # ── Qwen 2.5 ──
-  - name: "qwen-2.5-72b"
-    routes:
-      - provider: groq
-        model: "qwen-2.5-72b"
-      - provider: deepinfra
-        model: "Qwen/Qwen2.5-72B-Instruct"
-        pricing: { input: 0.23, output: 0.40 }
-
-  - name: "qwen-2.5-coder-32b"
-    routes:
-      - provider: groq
-        model: "qwen-2.5-coder-32b"
-      - provider: deepinfra
-        model: "Qwen/Qwen2.5-Coder-32B-Instruct"
-        pricing: { input: 0.07, output: 0.16 }
-
-  # ── Other ──
-  - name: "gemma-2-9b"
-    routes:
-      - provider: groq
-        model: "gemma2-9b-it"
-
-  - name: "deepseek-r1-distill-llama-70b"
-    routes:
-      - provider: groq
-        model: "deepseek-r1-distill-llama-70b"
-      - provider: deepinfra
-        model: "deepseek-ai/DeepSeek-R1-Distill-Llama-70B"
-        pricing: { input: 0.23, output: 0.69 }
-
-  - name: "deepseek-r1-distill-qwen-32b"
-    routes:
-      - provider: deepinfra
-        model: "deepseek-ai/DeepSeek-R1-Distill-Qwen-32B"
-        pricing: { input: 0.07, output: 0.16 }
-
-  # ── OpenRouter premium ──
-  - name: "deepseek-v3-free"
-    routes:
-      - provider: openrouter
-        model: "deepseek/deepseek-chat-v3-0324:free"
-
   - name: "minimax-m2.5"
     routes:
       - provider: openrouter
@@ -257,8 +273,3 @@ models:
     routes:
       - provider: openrouter
         model: "anthropic/claude-sonnet-4"
-
-  - name: "trinity-large-preview"
-    routes:
-      - provider: openrouter
-        model: "arcee-ai/trinity-large-preview"