BENCHMARK COLLECTIONS

Benchmark collections

Independent scoring references, excluded from the main timeline, collection and case total. Choose a benchmark, then inspect model summaries and original samples.

Upstream outputs, not a Pelican Map ranking

An index of upstream evaluation material, not a capability leaderboard. Source model labels are not independently authenticated.

nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4 · Original sample

2026-07-29 · 7 model labels · 139 main-task rows

HuggingFace / OpenEnv · Pelican SVG scoring reference

One upstream experiment: data published on HuggingFace, scored by the OpenEnv environment. Seven model labels and 139 main-task rows; 30 prompt-coverage samples are separate.

Open benchmark →