The Open-Source AI Stack
RSS
All models

Models · Compare

OpenAI o3 vs Llama-3.1-Nemotron Ultra 253B v1

Rows highlighted in warm gray are where the models differ. Numbers carry their as-of date and primary source.

Specs

Field A: OpenAI o3 B: Llama-3.1-Nemotron Ultra 253B v1
Released 2025-04-162025-04-11
Developer OpenAINVIDIA
Openness ProprietaryOpen weights
License ProprietaryNVIDIA Open Model License
OSI-approved nono
Data released noyes
Training code nono
Architecture unknowndense
Total params 253B
Active params
Experts
Context window 200K131K
Attention unknownskip-attention
Position enc. unknownrope
Pretraining tokens 65B
Post-training rlhfsft, grpo
Training hardware H100
$/M input $2.00
$/M output $8.00
Output tok/sec 88

Benchmarks

Missing scores render as not reported; never inferred. Bold highlights the leader per benchmark.

General reasoning

MMLU-Pro 85.3 2026-05-21
GPQA-Diamond 87.7 2025-04-16 76.0 2025-04-11

Code

SWE-Bench Verified 71.7 2025-04-16
LiveCodeBench 80.8 2026-05-21 66.3 2025-04-11

Math

MATH 99.2 2026-05-21 97.0 2025-04-11
AIME 2024 90.3 2026-05-21
AIME 2025 88.3 2026-05-21 72.5 2025-04-11

Held-out / arena

IFEval 88.8 2025-04-11

Context · A

Released April 16, 2025 as the full-size successor to o1 in the o-series reasoning lineage, with multimodal (text plus image) input and a 200K-token context. OpenAI reported large jumps over o1 on SWE-bench Verified and Codeforces and roughly 3x the accuracy of o1 on ARC-AGI.

Context · B

Top of NVIDIA's Llama-Nemotron family, distilled from Llama 3.1 405B via Neural Architecture Search with skip attention, variable FFN, and FFN fusion. Released April 11 2025; single-node 8x H100 BF16 inference, 4x H100 FP8. Post-trained through SFT and GRPO RL stages.

OpenAI o3 detail → · Llama-3.1-Nemotron Ultra 253B v1 detail →