HELM SeaHELM - TyDiQA — leaderboard

Metric: SQuAD macro-averaged F1 score. Source: crfm.stanford.edu. 21 models tracked.

Top models

#ModelScore
1Gemini 2.0 Flash81.51
2Gemini 2.0 Flash Lite65.01
3Llama 4 Maverick Instruct FP864.9
4GPT-4o (2024-11-20)63.68
5Gemma 2 9B (IT)62.17
6Llama 3.2 90B Vision Instruct60.07
7Llama 3.3 70B Instruct57.34
8Gemma 2 27B (IT)56.32
9Llama 4 Scout Instruct55.91
10GPT-4o Mini (2024-07-18)52.46
11Llama 3.1 405B Instruct52.2
12Llama 3.1 70B Instruct51.04
13Qwen 2 72B Instruct50.52
14DeepSeek R1 Distill Llama 8B50.19
15Qwen 2.5 72B Instruct40.7

Interactive version: theaggregate.ai/benchmark?slug=helm-seahelm-tydiqa · How the rankings work · Data refreshed daily, snapshot 2026-07-22.