Skip to main content
BenchLM
Advertise

Model comparison

GPT-5.4 mini vs Nemotron 3 Ultra

Data verified

Head-to-head evidence from 20 shared benchmark results across 5 categories. Overall scores shown here use the public BenchAlign v5 ranking lane.

56.69/100
Margin
3.3pts
winning →
60/100
1 category wins1 category wins

Verified leaderboard positions: GPT-5.4 mini unranked; Nemotron 3 Ultra #24

BenchAlign evidence: GPT-5.4 mini estimated; Nemotron 3 Ultra not scored. Intervals and evidence labels describe ranking uncertainty, not a guarantee for a specific workload.

Evidence parity. GPT-5.4 mini and Nemotron 3 Ultra share 20 comparable benchmark results. 2 of 8 categories are comparable. 11 results are unique to GPT-5.4 mini; 19 to Nemotron 3 Ultra.

Updated July 16, 2026
Shared results
20
GPT-5.4 mini only
11
Nemotron 3 Ultra only
19
Comparable categories
2 / 8

Pick GPT-5.4 mini if you want the stronger benchmark profile. Nemotron 3 Ultra only becomes the better choice if knowledge is the priority or you want the cheaper token bill.

Confidence note. This is a partial-evidence comparison with 20 shared benchmark results across 5 evidence categories; 2 of 8 categories currently have scoreable aggregates for both models. Treat the verdict as directional until coverage is more balanced.

Why this result

GPT-5.4 mini is clearly ahead on the provisional aggregate, 64 to 60. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.

GPT-5.4 mini's sharpest advantage is in agentic, where it averages 65.7 against 51.3. The single biggest benchmark swing on the page is HLE, 41.5% to 26.7%. Nemotron 3 Ultra does hit back in knowledge, so the answer changes if that is the part of the workload you care about most.

GPT-5.4 mini is also the more expensive model on tokens at $0.75 input / $4.50 output per 1M tokens, versus $0.00 input / $0.00 output per 1M tokens for Nemotron 3 Ultra. That is roughly Infinityx on output cost alone. Nemotron 3 Ultra gives you the larger context window at 1M, compared with 400K for GPT-5.4 mini.

Category breakdown

Exact category averages are shown below. Not measured means BenchLM does not have enough sourced public coverage for that model and category.

Category scores and score margins for GPT-5.4 mini and Nemotron 3 Ultra
CategoryGPT-5.4 miniΔNemotron 3 Ultra
AgenticGPT-5.4 mini65.7Margin← 14.4Nemotron 3 Ultra51.3
KnowledgeGPT-5.4 mini47.8Margin→ 6.4Nemotron 3 Ultra54.2
CodingGPT-5.4 miniNot measuredMarginNo overlapNemotron 3 Ultra58.3
ReasoningGPT-5.4 miniNot measuredMarginNo overlapNemotron 3 Ultra61.9
MathGPT-5.4 mini21.7MarginNo overlapNemotron 3 UltraNot measured
MultilingualGPT-5.4 miniNot measuredMarginNo overlapNemotron 3 Ultra83.0
MultimodalGPT-5.4 mini76.6MarginNo overlapNemotron 3 UltraNot measured
Inst. FollowingGPT-5.4 miniNot measuredMarginNo overlapNemotron 3 Ultra81.7

Decisive benchmark drivers

The largest measured benchmark gaps in this matchup, with exact reported values.

More
A · GPT-5.4 miniB · Nemotron 3 Ultra

  1. HLE

    Knowledge
    Source ↗
    A 41.5%B 26.7%
    Winner: GPT-5.4 miniΔ 14.8
    HLE: GPT-5.4 mini scored 41.5%; Nemotron 3 Ultra scored 26.7%. GPT-5.4 mini wins this benchmark.

  2. Terminal-Bench 2.0

    Agentic
    Source ↗
    A 60%B 56.4%
    Winner: GPT-5.4 miniΔ 3.6
    Terminal-Bench 2.0: GPT-5.4 mini scored 60%; Nemotron 3 Ultra scored 56.4%. GPT-5.4 mini wins this benchmark.

  3. GPQA

    Knowledge
    Source ↗
    A 88%B 87%
    Winner: GPT-5.4 miniΔ 1
    GPQA: GPT-5.4 mini scored 88%; Nemotron 3 Ultra scored 87%. GPT-5.4 mini wins this benchmark.

Operational comparison

Runtime and commercial metrics are compared only when both models have a complete sourced value.

MetricGPT-5.4 miniNemotron 3 UltraComparison
Input / output priceUSD per 1M tokensGPT-5.4 mini$0.75 input / $4.5 outputNemotron 3 Ultra$0 input / $0 outputNemotron 3 Ultra has the lower combined listed price.
Generation speedtokens per secondGPT-5.4 mini201 tok/sNemotron 3 UltraNot availableA complete speed comparison is not available.
First-answer latencyseconds to first tokenGPT-5.4 mini3.85 sNemotron 3 UltraNot availableA complete latency comparison is not available.
Context windowmaximum listed tokensGPT-5.4 mini400KNemotron 3 Ultra1MNemotron 3 Ultra lists the larger context window.

Benchmark Deep Dive

AgenticGPT-5.4 mini wins
BenchmarkGPT-5.4 miniNemotron 3 UltraResult
Terminal-Bench 2.0Source 60%56.4%GPT-5.4 mini leads
OSWorld-VerifiedSource 72.1%—Not comparable
MCP AtlasSource 57.7%—Not comparable
ToolathlonSource 42.9%—Not comparable
τ²-bench resultsSource 83.3%83.3%Tie
AA Agentic IndexSource 30.2%27.4%GPT-5.4 mini leads
APEX-Agents-AASource 28.2%—Not comparable
GDPval-AASource 33.5%33.2%GPT-5.4 mini leads
GDPval-AASource 11711164GPT-5.4 mini leads
PinchBenchSource —90.0%Not comparable
BrowseCompSource —44.4%Not comparable
τ³-bench resultsSource —70.9%Not comparable
HLE w/ toolsSource —37.4%Not comparable
AA BriefcaseSource —870Not comparable
AA EnterpriseOps-GymSource —28.9%Not comparable
AA Harvey LABSource —3.3%Not comparable
Coding
BenchmarkGPT-5.4 miniNemotron 3 UltraResult
Vibe Code BenchSource 47.97%—Not comparable
AA Coding IndexSource 56.1%49.3%GPT-5.4 mini leads
Terminal-Bench HardSource 52.3%36.4%GPT-5.4 mini leads
AA-SciCodeSource 49.9%39.9%GPT-5.4 mini leads
FrontierCodeSource 27.0%—Not comparable
SWE-bench VerifiedSource —71.9%Not comparable
SWE MultilingualSource —67.7%Not comparable
LiveCodeBench v6Source —89.0%Not comparable
SciCodeSource —44.6%Not comparable
Terminal-Bench 2.0Source —56.4%Not comparable
Reasoning
BenchmarkGPT-5.4 miniNemotron 3 UltraResult
AA-LCRSource 69.3%67.0%GPT-5.4 mini leads
CritPtSource 10.0%3.1%GPT-5.4 mini leads
LongBench v2Source —61.9%Not comparable
KnowledgeNemotron 3 Ultra wins
BenchmarkGPT-5.4 miniNemotron 3 UltraResult
GPQASource 88%87%GPT-5.4 mini leads
HLESource 41.5%26.7%GPT-5.4 mini leads
HLE w/o toolsSource 28.2%26.7%GPT-5.4 mini leads
Artificial Analysis Intelligence IndexSource 40.0%37.8%GPT-5.4 mini leads
AA-GPQA DiamondSource 87.5%86.7%GPT-5.4 mini leads
AA-HLESource 26.6%26.6%Tie
AA-Omniscience IndexSource -18.7%-0.8%Nemotron 3 Ultra leads
AA-Omniscience AccuracySource 37.5%21.6%GPT-5.4 mini leads
AA-Omniscience Hallucination RateSource 89.8%28.5%Nemotron 3 Ultra leads
GPQA-DSource —87.0%Not comparable
MMLU-ProSource —86.8%Not comparable
AA Openness IndexSource —83.3%Not comparable
Math
BenchmarkGPT-5.4 miniNemotron 3 UltraResult
FrontierMath v2 (Tiers 1-3)Source 28.280%—Not comparable
FrontierMath v2 (Tier 4)Source 2.080%—Not comparable
Multilingual
BenchmarkGPT-5.4 miniNemotron 3 UltraResult
MMLU-ProXSource —83%Not comparable
Multimodal
BenchmarkGPT-5.4 miniNemotron 3 UltraResult
MMMU-ProSource 76.6%—Not comparable
MMMU-Pro w/ PythonSource 78%—Not comparable
AA-MMMU-ProSource 73.3%—Not comparable
Design Arena WebsiteSource —1132Not comparable
Inst. Following
BenchmarkGPT-5.4 miniNemotron 3 UltraResult
AA-IFBenchSource 73.3%81.4%Nemotron 3 Ultra leads
IFBenchSource —81.7%Not comparable
Frequently Asked Questions (3)

Which is better, GPT-5.4 mini or Nemotron 3 Ultra?

GPT-5.4 mini is ahead on BenchLM's provisional leaderboard, 64 to 60. The biggest single separator in this matchup is HLE, where the scores are 41.5% and 26.7%.

Which is better for knowledge tasks, GPT-5.4 mini or Nemotron 3 Ultra?

Nemotron 3 Ultra has the edge for knowledge tasks in this comparison, averaging 54.2 versus 47.8. Inside this category, AA-Omniscience Hallucination Rate is the benchmark that creates the most daylight between them.

Which is better for agentic tasks, GPT-5.4 mini or Nemotron 3 Ultra?

GPT-5.4 mini has the edge for agentic tasks in this comparison, averaging 65.7 versus 51.3. Inside this category, GDPval-AA is the benchmark that creates the most daylight between them.

Related Comparisons

Explore More

OpenAI Compare Pricing Methodology Find Your Best LLM Overall Rankings
Last updated: July 16, 2026

The AI models change fast. We track them for you.

A weekly brief for engineers and researchers covering new models, ranking shifts, and pricing changes.

Free. No spam. Unsubscribe anytime.