K2 Horizon FTPO Card - HMMT Feb 2026: leaderboard

Competition problems from the February 2026 Harvard-MIT Mathematics Tournament. Scores from IFM's K2 Horizon 3.7B model card at revision 4525090a, whose anti-doom-loop table evaluates the 3.7B and 7B launch checkpoints and the checkpoints IFM promoted on 1 October 2026 after Final Token Preference Optimization (FTPO), with identical settings. Output budget 256k tokens. Metric: accuracy (%).

Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Around February 2027. 4 models tracked.

Top models

#ModelScoreOverall rank
1K2 Horizon 7B73.3#714
2K2 Horizon 3.7B69.3#1075

No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.

Interactive version: theaggregate.ai/benchmark?slug=k2-horizon-ftpo-card-hmmt-feb-2026 · How It Works · Data refreshed daily, snapshot 2026-11-12.