HMMT25
A reasoning benchmark composed of problems from the 2025 Harvard-MIT Mathematics Tournament (HMMT), used to evaluate LLMs on competition-level mathematics tasks, often in conjunction with the avg@12 metric.
Timeline (last 90 days)
In the Qwen3-4B comparative experiment, updating the model directly with unfiltered skills without using SGUID filtering decreased performance by 0.3 percentage points on HMMT25.
When using SGUID, Qwen3-4B improved by 0.5 percentage points on HMMT25 after the first round, and by a cumulative 1.1 percentage points after the second round.
All Facts (2)
In the Qwen3-4B comparative experiment, updating the model directly with unfiltered skills without using SGUID filtering decreased performance by 0.3 percentage points on HMMT25.
50%UnverifiedWhen using SGUID, Qwen3-4B improved by 0.5 percentage points on HMMT25 after the first round, and by a cumulative 1.1 percentage points after the second round.
50%