Llama 3.1 Nemotron Ultra 253B v1 (Reasoning)
NVIDIALlama開源權重Llama 3.1 Community License
描述
A 253B parameter derivative of Meta Llama 3.1 405B Instruct, developed by NVIDIA using Neural Architecture Search (NAS) and vertical compression. It underwent multi-phase post-training (SFT for Math, Code, Reasoning, Chat, Tool Calling; RL with GRPO) to enhance reasoning and instruction-following. Optimized for accuracy/efficiency tradeoff on NVIDIA GPUs. Supports 128k context.
發布日期
2025-04-07
參數規模
253.0B
上下文長度
128K
支援模態
text
能力雷達圖
30
general
64
coding
72
reasoning
53
science
70
agents
0
multimodal
排行榜排名
基準測試分數 (LLM Stats)
(LLM Stats (zeroeval))Chat
IFEvalGoogle Research (2023)
89.5%自報
General
BFCL v2
74.1%自報
Math
MATH-500
97.0%自報
AIME 2025
72.5%自報
Reasoning
GPQANYU + Cohere + Anthropic (2023)
76.0%自報
LiveCodeBench
66.3%自報
AA 評測指數
(Artificial Analysis)Math 500(OpenAI (2024), subset of Hendrycks et al. MATH (2021))95.2
Mmlu Pro(TIGER-Lab (Univ. of Waterloo, Toronto, CMU, 2024))82.5
Aime(MAA (Mathematical Association of America))74.7
Gpqa(NYU + Cohere + Anthropic (2023))72.8
Livecodebench(UC Berkeley + MIT + Cornell (2024))64.1
Math Index(Artificial Analysis)63.7
Aime 25(MAA (Mathematical Association of America))63.7
Ifbench(Google Research (2023))38.2
Tau2(Sierra + U Toronto + Vector Institute (2025))11.4
Intelligence Index(Artificial Analysis)7.5
Hle(Center for AI Safety + Scale AI (2025))7.4
Terminalbench Hard(Stanford × Laude Institute (2026))2.3
LLM Stats 分類評分
(LLM Stats (zeroeval))Chat90
Instruction Following90
Structured Output90
Math80
Physics80
Reasoning80
General80
Biology80
Chemistry80
Code70
Tool Calling70
定價
輸入價格免費
輸出價格免費
混合價格(3:1)免費
速度
Tokens/秒0.0
首Token延遲0.00s
首回答延遲0.00s
供應商價格排行
暫無提供商資料