Longbench v2 (≤ 1M)

Longbench v2 (≤ 1M) - long context understanding benchmark, up to 1M tokens.

Category

Long Context

Max Score

100.0

Score Type

percent

Active

Yes

Model Scores

Scores for Longbench v2 (≤ 1M)

Model Score Date Verified
NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16
41.18%
04.06.2026 Verified
Qwen3.5-397B-A17B
100.00%
04.06.2026 Unverified
DeepSeek-V4-Pro
42.86%
04.06.2026 Unverified
DeepSeek-V4-Flash
57.00
04.06.2026 Unverified