Cheaper overall
GLM 4.5 Air
8% lower blended rate
Higher benchmark scores
GPT-5.4 Nano
leads on 9 of 9 figures
Bigger context window
GPT-5.4 Nano
400,000 vs 131,072 tokens
Cheaper cached input
GPT-5.4 Nano
$0.02 vs $0.03, 33% less
Where each one wins
GPT-5.4 Nano
- Cheaper cached input $0.02 vs $0.03, 33% less
- Larger context window 400,000 tokens
- Longer maximum output 128,000 tokens
- Higher graduate-level science (GPQA Diamond) 81.7 vs 73.3
- Higher expert-exam performance (Humanity's Last Exam) 28.3 vs 7
- Higher instruction following (IFBench) 75.9 vs 37.6
- Ahead on 6 more benchmark figures
- Offers Batch and Flex pricing not listed for GLM 4.5 Air
GLM 4.5 Air
- Cheaper output tokens $1.10 vs $1.25, 12% less
What four workloads cost
One run, and the same run a thousand times.
Caching is what decides it: over twenty calls GPT-5.4 Nano costs $0.0375 against $0.0414, a difference of $3.90 across a thousand loops.
| Workload | GPT-5.4 Nano | GLM 4.5 Air | ร1,000 | Difference |
|---|---|---|---|---|
| Chat turn 1,000 in / 500 out | $0.0008 | $0.0008 | $0.8250 / $0.7500 | GLM 4.5 Air is 9% cheaper |
| RAG answer 10,000 in / 800 out | $0.0030 | $0.0029 | $3.00 / $2.88 | GLM 4.5 Air is 4% cheaper |
| Agent loop 32,000 in / 700 out ร 20 calls | $0.0375 | $0.0414 | $37.50 / $41.40 | GPT-5.4 Nano is 9% cheaper |
| Long document 400,000 in / 3,000 out | GLM 4.5 Air's context window holds 131,072 tokens, so a 400,000-token prompt does not fit. | |||
Price per 1M tokens
Above 272,000 input tokens the rates change, and the new rate applies to the whole request, not just the tokens past that point. It does not change which of the two is cheaper.
| Context band | Price | GPT-5.4 Nano | GLM 4.5 Air | Difference |
|---|---|---|---|---|
| Prompts up to 272,000 tokens | Input | $0.20 | $0.20 | Same |
| Prompts up to 272,000 tokens | Cached input | $0.02 | $0.03 | GPT-5.4 Nano is 33% cheaper |
| Prompts up to 272,000 tokens | Output | $1.25 | $1.10 | GLM 4.5 Air is 12% cheaper |
| Prompts over 272,000 tokens | Input | $0.20 | $0.20 | Same |
| Prompts over 272,000 tokens | Cached input | $0.02 | $0.03 | GPT-5.4 Nano is 33% cheaper |
| Prompts over 272,000 tokens | Output | $1.25 | $1.10 | GLM 4.5 Air is 12% cheaper |
Blended: GPT-5.4 Nano $0.4625, GLM 4.5 Air $0.425 per 1M โ one rate at 3:1 input to output, for comparing two models at a glance.
GPT-5.4 Nano priced by OpenAI, GLM 4.5 Air by Z.ai. Standard tier, pay-as-you-go.
Other pricing tiers
Only GPT-5.4 Nano lists Batch and Flex, at up to 50% off its own standard rate; GLM 4.5 Air 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 โ how we read tiers.
| Tier | GPT-5.4 Nano | GLM 4.5 Air | Cheaper |
|---|---|---|---|
| Batch only on GPT-5.4 Nano | $0.10 in / $0.625 out 50% off Standard | Not offered | — |
| Flex only on GPT-5.4 Nano | $0.10 in / $0.625 out 50% off Standard | Not offered | — |
Benchmarks
GPT-5.4 Nano leads on all 9 figures, furthest ahead on answer reliability (Omniscience Non-Hallucination), by 41.8 points.
Graduate-level science
GPQA Diamond โ Graduate-level questions in biology, chemistry & physics
GPT-5.4 Nano ahead by 8.4
Expert-exam performance
Humanity's Last Exam โ Expert-level questions across many academic domains
GPT-5.4 Nano ahead by 21.3
Instruction following
IFBench โ Precise following of detailed instructions
GPT-5.4 Nano ahead by 38.3
Support-agent performance
ฯยฒ-Bench Telecom โ Tool-using agent tasks in a telecom support setting
GPT-5.4 Nano ahead by 29.5
Long-context reasoning
AA-LCR โ Long-context reasoning across large inputs
GPT-5.4 Nano ahead by 30
Command-line work
Terminal-Bench Hard โ Complex command-line and terminal workflows
GPT-5.4 Nano ahead by 21.9
Physics research reasoning
CritPt โ Unpublished physics research reasoning problems
GPT-5.4 Nano ahead by 9.3
Factual accuracy
Omniscience Accuracy โ Breadth of factual knowledge across domains
GPT-5.4 Nano ahead by 9.4
Answer reliability
Omniscience Non-Hallucination โ How reliably the model avoids fabricated answers
GPT-5.4 Nano ahead by 41.8
Each figure is that model's best published run, with the effort level named beside it โ where these scores come from.
Specifications
GPT-5.4 Nano holds 268,928 more input tokens in one request. GPT-5.4 Nano also takes file and image.
| Specification | GPT-5.4 Nano | GLM 4.5 Air |
|---|---|---|
| Context window | 400,000 tokens | 131,072 tokens |
| Max output | 128,000 tokens | 98,304 tokens |
| Takes in | Text, image, file | Text |
| Puts out | Text | Text |
| Knowledge cutoff | 31 Aug 2025 | 31 Dec 2024 |
| Released | 17 Mar 2026 | 26 Jul 2025 |
| Status | Active | Active |
| Sold by | OpenAI, Perplexity | Z.ai |
FAQs
- Is GPT-5.4 Nano cheaper than GLM 4.5 Air?
- No โ the other way round. A 1,000-token prompt with a 500-token reply costs $0.0008 on GPT-5.4 Nano and $0.0008 on GLM 4.5 Air.
- How much do GPT-5.4 Nano and GLM 4.5 Air cost per 1M tokens?
- GPT-5.4 Nano costs $0.20 for input and $1.25 for output. GLM 4.5 Air costs $0.20 and $1.10. Standard pay-as-you-go rates, as of 26 Jul 2025.
- Which is better, GPT-5.4 Nano or GLM 4.5 Air?
- On published benchmarks, GPT-5.4 Nano. It leads on 9 of the 9 figures both models report, including graduate-level science (GPQA Diamond), where it scores 81.7 against 73.3.
- Does GPT-5.4 Nano or GLM 4.5 Air offer batch pricing?
- GPT-5.4 Nano only. Its batch input costs $0.10, 50% off its standard rate, and GLM 4.5 Air lists no batch tier at all.
- Does GPT-5.4 Nano or GLM 4.5 Air have a bigger context window?
- GPT-5.4 Nano, at 400,000 tokens against 131,072. That is 268,928 more input tokens in a single request.