diff --git a/litellm/config.yaml b/litellm/config.yaml index 6fe6f98..a7a78ec 100644 --- a/litellm/config.yaml +++ b/litellm/config.yaml @@ -22,23 +22,51 @@ model_list: api_key: os.environ/OPENROUTER_API_KEY # ═══════════════════════════════════════════════ - # TIER 2: SiliconFlow (cheapest paid, ~3-5x cheaper than OpenRouter) + # TIER 2: DeepSeek V3.2 (cheapest first) # ═══════════════════════════════════════════════ - # DeepSeek V3 — best value daily driver ($0.13 in / $0.28 out per M) + # DeepSeek V3.2 via DeepInfra ($0.26 in / $0.38 out per M) - model_name: deepseek-v3 litellm_params: - model: openai/deepseek-ai/DeepSeek-V3-0324 + model: deepinfra/deepseek-ai/DeepSeek-V3.2 + api_key: os.environ/DEEPINFRA_API_KEY + + - model_name: deepseek-v3.2 + litellm_params: + model: deepinfra/deepseek-ai/DeepSeek-V3.2 + api_key: os.environ/DEEPINFRA_API_KEY + + # DeepSeek V3.2 fallback via SiliconFlow ($0.27 in / $0.42 out per M) + - model_name: deepseek-v3 + litellm_params: + model: openai/deepseek-ai/DeepSeek-V3.2 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 + model: openai/deepseek-ai/DeepSeek-V3.2 api_base: https://api.siliconflow.com/v1 api_key: os.environ/SILICONFLOW_API_KEY + # ═══════════════════════════════════════════════ + # TIER 3: Other DeepInfra models + # ═══════════════════════════════════════════════ + + - 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: SiliconFlow (Qwen/GLM) + # ═══════════════════════════════════════════════ + # GLM-4.7 via SiliconFlow - model_name: glm-4.7 litellm_params: @@ -61,27 +89,7 @@ model_list: 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) + # TIER 5: OpenRouter (most expensive, widest selection) # ═══════════════════════════════════════════════ # Kimi K2.5 — DeepInfra is cheapest ($0.45 in / $2.25 out per M)