diff --git a/llm-gateway.yaml b/llm-gateway.yaml index 51b23d5..b6a897b 100644 --- a/llm-gateway.yaml +++ b/llm-gateway.yaml @@ -314,6 +314,58 @@ models: - provider: openrouter model: "google/gemini-2.5-pro-preview" + # ═══ TIER 10: Vision / Multimodal ═══ + - name: "gemma-3-4b" + routes: + - provider: openrouter + model: "google/gemma-3-4b-it" + pricing: { input: 0.017, output: 0.068 } + - provider: deepinfra + model: "google/gemma-3-4b-it" + pricing: { input: 0.04, output: 0.08 } + + - name: "gemma-3-12b" + routes: + - provider: openrouter + model: "google/gemma-3-12b-it" + pricing: { input: 0.03, output: 0.10 } + - provider: deepinfra + model: "google/gemma-3-12b-it" + pricing: { input: 0.04, output: 0.13 } + + - name: "gemma-3-27b" + routes: + - provider: openrouter + model: "google/gemma-3-27b-it" + pricing: { input: 0.04, output: 0.15 } + - provider: deepinfra + model: "google/gemma-3-27b-it" + pricing: { input: 0.08, output: 0.16 } + + - name: "qwen3-vl-8b" + routes: + - provider: openrouter + model: "qwen/qwen3-vl-8b-instruct" + pricing: { input: 0.08, output: 0.50 } + - provider: deepinfra + model: "Qwen/Qwen3-VL-8B-Instruct" + pricing: { input: 0.18, output: 0.69 } + + - name: "qwen3-vl-32b" + routes: + - provider: openrouter + model: "qwen/qwen3-vl-32b-instruct" + pricing: { input: 0.104, output: 0.416 } + + - name: "qwen2.5-vl-32b" + routes: + - provider: openrouter + model: "qwen/qwen2.5-vl-32b-instruct" + pricing: { input: 0.05, output: 0.22 } + - provider: deepinfra + model: "Qwen/Qwen2.5-VL-32B-Instruct" + pricing: { input: 0.20, output: 0.60 } + - name: "claude-sonnet" routes: - provider: openrouter