Portkey AI is currently processing over 12 billion large language model (LLM) tokens daily through its AI Gateway. The company has analyzed the performance of its top models by mapping LMSys scores against output token costs. Additionally, new insights into model efficiency have emerged from the Open LLM Leaderboard, which now includes CO₂ emissions data for each model evaluation. This feature allows users to assess the environmental impact of models, highlighting which ones offer the best CO₂ inference cost to performance ratio. Notably, models from Alibaba Qwen have been identified as performing well in this regard. The leaderboard also emphasizes that higher costs do not necessarily correlate with better performance, as smaller models may have unpredictable inference costs.