| model_cost = { | |
| "deepseek-ai/DeepSeek-R1": { | |
| "input_token_cost": 4, | |
| "output_token_cost": 16, | |
| }, | |
| "deepseek-ai/DeepSeek-V3": { | |
| "input_token_cost": 2, | |
| "output_token_cost": 8, | |
| }, | |
| "Pro/deepseek-ai/DeepSeek-R1":{ | |
| "input_token_cost": 4, | |
| "output_token_cost": 16, | |
| }, | |
| "Pro/deepseek-ai/DeepSeek-V3": { | |
| "input_token_cost": 2, | |
| "output_token_cost": 8, | |
| }, | |
| "deepseek-ai/DeepSeek-R1-Distill-Llama-70B": { | |
| "token_cost": 4.13 | |
| }, | |
| "deepseek-ai/DeepSeek-R1-Distill-Qwen-14B": { | |
| "token_cost": 0.7 | |
| }, | |
| "Qwen/QVQ-72B-Preview": { | |
| "token_cost": 9.9 | |
| }, | |
| "deepseek-ai/DeepSeek-V2.5": { | |
| "token_cost": 1.33 | |
| }, | |
| "meta-llama/Llama-3.3-70B-Instruct": { | |
| "token_cost": 4.13, | |
| }, | |
| "Qwen/QwQ-32B-Preview": { | |
| "token_cost": 1.26 | |
| }, | |
| "Qwen/Qwen2.5-Coder-32B-Instruct": { | |
| "token_cost": 1.26 | |
| }, | |
| "Qwen/Qwen2-VL-72B-Instruct": { | |
| "token_cost": 4.13 | |
| }, | |
| "OpenGVLab/InternVL2-26B": { | |
| "token_cost": 1 | |
| }, | |
| "Qwen/Qwen2.5-72B-Instruct-128K": { | |
| "token_cost": 4.13 | |
| }, | |
| "deepseek-ai/deepseek-vl2": { | |
| "token_cost": 0.99 | |
| }, | |
| "Qwen/Qwen2.5-72B-Instruct": { | |
| "token_cost": 4.13 | |
| }, | |
| "Qwen/Qwen2.5-32B-Instruct": { | |
| "token_cost": 1.26 | |
| }, | |
| "Qwen/Qwen2.5-14B-Instruct": { | |
| "token_cost": 0.7 | |
| }, | |
| "TeleAI/TeleChat2": { | |
| "token_cost": 1.33 | |
| }, | |
| "internlm/internlm2_5-20b-chat": { | |
| "token_cost": 1 | |
| }, | |
| "meta-llama/Meta-Llama-3.1-405B-Instruct": { | |
| "token_cost": 21 | |
| }, | |
| "meta-llama/Meta-Llama-3.1-70B-Instruct": { | |
| "token_cost": 4.13 | |
| }, | |
| "01-ai/Yi-1.5-34B-Chat-16K": { | |
| "token_cost": 1.26 | |
| }, | |
| "google/gemma-2-27b-it": { | |
| "token_cost": 1.26 | |
| }, | |
| "LoRA/meta-llama/Meta-Llama-3.1-8B-Instruct": { | |
| "token_cost": 0.63 | |
| }, | |
| "LoRA/Qwen/Qwen2.5-32B-Instruct": { | |
| "token_cost": 0.63 | |
| }, | |
| "LoRA/Qwen/Qwen2.5-14B-Instruct": { | |
| "token_cost": 1.05 | |
| }, | |
| "Vendor-A/Qwen/Qwen2.5-72B-Instruct": { | |
| "token_cost": 1 | |
| }, | |
| "Pro/deepseek-ai/DeepSeek-R1-Distill-Llama-8B": { | |
| "token_cost": 0.42 | |
| }, | |
| "Pro/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B": { | |
| "token_cost": 0.35 | |
| }, | |
| "Pro/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B": { | |
| "token_cost": 0.14 | |
| }, | |
| "Pro/Qwen/Qwen2.5-Coder-7B-Instruct": { | |
| "token_cost": 0.35 | |
| }, | |
| "Pro/Qwen/Qwen2-VL-7B-Instruct": { | |
| "token_cost": 0.35 | |
| }, | |
| "Pro/OpenGVLab/InternVL2-8B": { | |
| "token_cost": 0.35 | |
| }, | |
| "Pro/Qwen/Qwen2.5-7B-Instruct": { | |
| "token_cost": 0.35 | |
| }, | |
| "Pro/meta-llama/Meta-Llama-3.1-8B-Instruct": { | |
| "token_cost": 0.42 | |
| }, | |
| "LoRA/Qwen/Qwen2.5-72B-Instruct": { | |
| "token_cost": 6.2 | |
| }, | |
| "Pro/Qwen/Qwen2-7B-Instruct": { | |
| "token_cost": 0.35 | |
| }, | |
| "Pro/Qwen/Qwen2-1.5B-Instruct": { | |
| "token_cost": 0.14 | |
| }, | |
| "LoRA/Qwen/Qwen2.5-7B-Instruct": { | |
| "token_cost": 0.53 | |
| }, | |
| "Pro/THUDM/glm-4-9b-chat": { | |
| "token_cost": 0.6 | |
| }, | |
| "Pro/google/gemma-2-9b-it": { | |
| "token_cost": 0.6 | |
| } | |
| } |