Cheaper overall
Nemotron-3.5 Super-120b-a12b
84% lower blended rate
Higher benchmark scores
Nemotron-3.5 Super-120b-a12b
leads on 5 of 8 figures
Bigger context window
Nemotron-3.5 Super-120b-a12b
1,048,576 vs 400,000 tokens
Cheaper cached input
Nemotron-3.5 Super-120b-a12b
$0.04 vs $0.075, 47% less
Where each one wins
Nemotron-3.5 Super-120b-a12b
- Cheaper input tokens $0.20 vs $0.75, 73% less
- Cheaper output tokens $0.50 vs $4.50, 89% less
- Cheaper cached input $0.04 vs $0.075, 47% less
- Cheaper on all four workloads
- Larger context window 1,048,576 tokens
- Longer maximum output 262,144 tokens
- Higher overall intelligence (Intelligence Index) 26.5 vs 24.1
- Higher expert-exam performance (Humanity's Last Exam) 30.7 vs 28.1
- Higher physics research reasoning (CritPt) 10.6 vs 10
- Ahead on 2 more benchmark figures
GPT-5.4 Mini
- Higher long-context reasoning (AA-LCR) 77 vs 69
- Higher scientific coding (SciCode) 52.1 vs 47.8
- Higher factual accuracy (Omniscience Accuracy) 37.5 vs 23.3
- Offers Batch and Flex pricing not listed for Nemotron-3.5 Super-120b-a12b
What four workloads cost
One run, and the same run a thousand times.
Nemotron-3.5 Super-120b-a12b is cheaper on all four workloads, by much the same margin each time — about 4.7 times, from $0.0005 against $0.0030 on a chat turn to $0.0815 against $0.3135 on a long document.
| Workload | Nemotron-3.5 Super-120b-a12b | GPT-5.4 Mini | ×1,000 | Difference |
|---|---|---|---|---|
| Chat turn 1,000 in / 500 out | $0.0005 | $0.0030 | $0.4500 / $3.00 | Nemotron-3.5 Super-120b-a12b is 85% cheaper |
| RAG answer 10,000 in / 800 out | $0.0024 | $0.0111 | $2.40 / $11.10 | Nemotron-3.5 Super-120b-a12b is 78% cheaper |
| Agent loop 32,000 in / 700 out × 20 calls | $0.0390 | $0.1380 | $39.00 / $138.00 | Nemotron-3.5 Super-120b-a12b is 72% cheaper |
| Long document 400,000 in / 3,000 out | $0.0815 | $0.3135 | $81.50 / $313.50 | Nemotron-3.5 Super-120b-a12b is 74% cheaper |
Price per 1M tokens
Neither model changes its rate with prompt length, so these rates apply to every request.
| Context band | Price | Nemotron-3.5 Super-120b-a12b | GPT-5.4 Mini | Difference |
|---|---|---|---|---|
| Any prompt length | Input | $0.20 | $0.75 | Nemotron-3.5 Super-120b-a12b is 73% cheaper |
| Any prompt length | Cached input | $0.04 | $0.075 | Nemotron-3.5 Super-120b-a12b is 47% cheaper |
| Any prompt length | Output | $0.50 | $4.50 | Nemotron-3.5 Super-120b-a12b is 89% cheaper |
Blended: Nemotron-3.5 Super-120b-a12b $0.275, GPT-5.4 Mini $1.6875 per 1M — one rate at 3:1 input to output, for comparing two models at a glance.
Nemotron-3.5 Super-120b-a12b priced by Perplexity, GPT-5.4 Mini by OpenAI. Standard tier, pay-as-you-go.
Other pricing tiers
Only GPT-5.4 Mini lists Batch and Flex, at up to 50% off its own standard rate; Nemotron-3.5 Super-120b-a12b offers none of them.
Every tier either model sells is here, so this is the whole price sheet. A saving is measured against that model's own standard rate — how we read tiers.
| Tier | Nemotron-3.5 Super-120b-a12b | GPT-5.4 Mini | Cheaper |
|---|---|---|---|
| Batch only on GPT-5.4 Mini | Not offered | $0.375 in / $2.25 out 50% off Standard | — |
| Flex only on GPT-5.4 Mini | Not offered | $0.375 in / $2.25 out 50% off Standard | — |
Benchmarks
They share 8 figures: Nemotron-3.5 Super-120b-a12b leads on 5, GPT-5.4 Mini on 3. The widest gap is 54 points, on answer reliability (Omniscience Non-Hallucination).
Overall intelligence
Intelligence Index — Overall intelligence across reasoning, knowledge & math evals
Nemotron-3.5 Super-120b-a12b ahead by 2.4
Expert-exam performance
Humanity's Last Exam — Expert-level questions across many academic domains
Nemotron-3.5 Super-120b-a12b ahead by 2.6
Long-context reasoning
AA-LCR — Long-context reasoning across large inputs
GPT-5.4 Mini ahead by 8
Scientific coding
SciCode — Research-level scientific coding tasks
GPT-5.4 Mini ahead by 4.3
Physics research reasoning
CritPt — Unpublished physics research reasoning problems
Nemotron-3.5 Super-120b-a12b ahead by 0.6
Real-world task performance
GDPval — Economically valuable, real-world knowledge work
Nemotron-3.5 Super-120b-a12b ahead by 7.4
Factual accuracy
Omniscience Accuracy — Breadth of factual knowledge across domains
GPT-5.4 Mini ahead by 14.2
Answer reliability
Omniscience Non-Hallucination — How reliably the model avoids fabricated answers
Nemotron-3.5 Super-120b-a12b ahead by 54
Each figure is that model's best published run, with the effort level named beside it — where these scores come from.
Specifications
Nemotron-3.5 Super-120b-a12b holds 648,576 more input tokens in one request. Nemotron-3.5 Super-120b-a12b also takes audio and video.
| Specification | Nemotron-3.5 Super-120b-a12b | GPT-5.4 Mini |
|---|---|---|
| Context window | 1,048,576 tokens | 400,000 tokens |
| Max output | 262,144 tokens | 128,000 tokens |
| Takes in | Text, audio, image, video, file | Text, image, file |
| Puts out | Text | Text |
| Knowledge cutoff | — | 31 Aug 2025 |
| Released | 15 Oct 2026 | 17 Mar 2026 |
| Status | Active | Active |
| Sold by | Perplexity | OpenAI |
FAQs
- Is Nemotron-3.5 Super-120b-a12b cheaper than GPT-5.4 Mini?
- Yes. A 1,000-token prompt with a 500-token reply costs $0.0005 on Nemotron-3.5 Super-120b-a12b and $0.0030 on GPT-5.4 Mini, and the same model is cheaper on every workload on this page.
- How much do Nemotron-3.5 Super-120b-a12b and GPT-5.4 Mini cost per 1M tokens?
- Nemotron-3.5 Super-120b-a12b costs $0.20 for input and $0.50 for output. GPT-5.4 Mini costs $0.75 and $4.50. Standard pay-as-you-go rates, as of 16 Oct 2026.
- Which is better, Nemotron-3.5 Super-120b-a12b or GPT-5.4 Mini?
- On published benchmarks, Nemotron-3.5 Super-120b-a12b. It leads on 5 of the 8 figures both models report, including overall intelligence (Intelligence Index), where it scores 26.5 against 24.1.
- Does Nemotron-3.5 Super-120b-a12b or GPT-5.4 Mini offer batch pricing?
- GPT-5.4 Mini only. Its batch input costs $0.375, 50% off its standard rate, and Nemotron-3.5 Super-120b-a12b lists no batch tier at all.
- Does Nemotron-3.5 Super-120b-a12b or GPT-5.4 Mini have a bigger context window?
- Nemotron-3.5 Super-120b-a12b, at 1,048,576 tokens against 400,000. That is 648,576 more input tokens in a single request.
Explore these models
Perplexity · 1,048,576-token context