The New Mercury 2.5 Model Can Reach Up To Nearly 800 TPS

Inception has launched their new model Mercury 2.5. According to our measurements on Artificial Analysis, it produces 781 tokens/second. This is almost seven times the median for models of its price ($0.25 per million tokens input, $0.75 per million tokens output), which is 110 tokens/second. The context size is 260 thousand tokens.

With such a fast model, you can create quick agentic loops, when the response needs to be provided immediately, such as voice assistants or code autocompletion. However, it scored only 12 points on the Artificial Analysis intelligence index and ranked below the median for models of a similar price, so I wouldn’t recommend using it instead of Opus 5 or Fable 5.1 for more complicated agentic tasks. Use fast and affordable models for simple tasks that require repetition, but don’t expect them to be able to handle complex reasoning.