See key specs and per-benchmark scores for each model/mode. Scroll horizontally for all columns. 当前对比 2 个模型的评测数据与核心参数。

DeepSeek-V4-Pro
DeepSeek-AI

Nemotron 3 Ultra
NVIDIA
Each axis is a category average, normalized to a 100-point radar.
Relative edge: AI Agent - 信息收集 +37.5 / Relative gap: 数学推理 -19.4
Relative edge: 数学推理 +19.4 / Relative gap: AI Agent - 信息收集 -37.5
Method: for each model and benchmark, the chart first averages all scores in the current mode scope instead of taking the best score, then averages those benchmark scores within each category. Only benchmarks with at least two selected models scored are included; missing values are not counted as zero.
Best overall
DeepSeek-V4-Pro · 80.08
Best single
DeepSeek-V4-Pro · LiveCodeBench 93.50
Modality coverage
DeepSeek-V4-Pro · 1 modalities
Head to head
9
Benchmarks
8
Wins
1
Losses
+13.80
Average diff
Compare benchmark results across thinking modes and tool usage.
Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology
Complete scores for each model/mode across selected benchmarks.
9 benchmarks with comparable scores. Each model shows its best score; mode label is displayed below.
| Benchmark | DeepSeek-V4-Pro | Nemotron 3 Ultra |
|---|---|---|
HLE 综合评估 | 48.20Thinking Level · Extra High | Tools | 37.40Thinking Enabled | Tools |
LiveBench 综合评估 | 73.58Standard Mode | 51.78Standard Mode |
MMLU Pro 综合评估 | 87.50Thinking Level · High | 86.80Thinking Enabled |
LiveCodeBench 编程与软件工程 | 93.50Thinking Level · High | 89.00Thinking Enabled |
SWE-bench Multilingual 编程与软件工程 | 76.20Thinking Level · Extra High | Tools | 67.70Thinking Enabled | Tools |
SWE-bench Verified 编程与软件工程 | 80.60Thinking Level · Extra High | Tools | 70.70Thinking Enabled | Tools |
BrowseComp AI Agent - 信息收集 | 83.40Thinking Level · Extra High | Tools | 44.40Thinking Enabled | Tools |
Terminal-Bench 2.1 AI Agent - 工具使用 | 87.90Thinking Level · Extra High | Tools | 56.40Thinking Enabled | Tools |
IMO-AnswerBench 数学推理 | 89.80Thinking Level · High | 92.30Thinking Enabled | Tools |
Side-by-side input/output token pricing
Licensing, MoE architecture, and multi-modality support.
| Features & specs | DeepSeek-V4-ProDeepSeek-AI | Nemotron 3 UltraNVIDIA |
|---|---|---|
Core specsRelease | 2026-08-13 | 2026-06-04 |
Context length | 1M | 1M |
Parameters | 16000 | 5500 |
Active parameters | 490 | 550 |
Max output | 384000 | Not provided |
MoE | Yes | Yes |
LicenseCode Open Source | Closed Source | Not provided |
Weights Open Source | Closed Source | Not provided |
Commercial use | 免费商用授权 | 免费商用授权 |
Modality supportText Input/Output | / | / |
ResourcesPaper / report | DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence | Nemotron 3 Ultra: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning |