DeepSeek 1017 checkpoint reportedly shows a small jump on FrontierCS 2

FrontierCS 2’s announcement says its tasks remove a paper’s contribution from released code and ask agents to beat the original method using the authors’ chosen metrics, without access to the paper or web. A user says the DeepSeek 1017 checkpoint is now on the benchmark with a small jump. Another commenter called the results “surprisingly poor” despite increased token consumption and asked whether they remained far below GLM 5.3.