TokenPost·

Modified DeepSeek Harness Uses 87.93% Fewer Tokens in Coding Test

Your call?

Its success rate also rose on the 98-task HumanEval benchmark. The token reduction reflects several changes to the agent setup.

Share

More news