Crosshair
Leaderboard
Language ModelsOpen weights

Nemotron 3 Ultra

NVIDIA's open-weights hybrid Mamba-MoE (550B total / 55B active), OpenMDW license.

Crosshair Index
63.8
#23 of 32 · Language Models
Provider
NVIDIA
Released
2026-06-04
Parameters
550B
Context
1M tokens

Token pricing

Input
$0.37 /1M
Output
$1.08 /1M

USD per 1M tokens · cache read = cached input (hit), cache write = caching surcharge · official list pricing (June 2026).

Industry skill web

Professional-domain strengths, composed from the benchmarks relevant to each field. Highlight an axis to see the benchmarks behind it.

SoftwareIB / FinanceLawMedicineResearchConsultingAccounting

Investment Banking

87
skill

Financial analysis over filings and credit agreements — valuation math, document QA, and the quantitative reasoning behind deals.

Scorecard

BenchmarkScoreSourceStatus
GPQA Diamond
Reasoning
87%NVIDIA — Nemotron 3 Ultra model cardvendorunverified
Humanity's Last Exam
Frontier
26.7%NVIDIA — Nemotron 3 Ultra model cardvendorunverified
SWE-bench Verified
Agentic Coding
71.9%NVIDIA — Nemotron 3 Ultra model cardvendorunverified
LiveCodeBench
Coding
89%NVIDIA — Nemotron 3 Ultra model cardvendorunverified
MMLU-Pro
Knowledge
86.8%NVIDIA — Nemotron 3 Ultra model cardvendorunverified
LMArena Elo
Human Preference
1,422LMArena (arena.ai)3rd-partyunverified
AA Intelligence Index
Composite
48Artificial Analysis3rd-partyunverified
Corporate Finance
Finance
not evaluated
LegalBench
Law
not evaluated
TaxEval
Tax & Accounting
not evaluated
Medical Coding
Medicine
not evaluated