OpenAI · Frontier
GPT 5.4
GPT-5.4 is an OpenAI model for professional work with stronger reasoning, coding, and tool use.
- Reasoning
- Tools
- Vision
- Structured output
- PDF input
- Input
- Text, Images
- Output
- Text
- Reasoning effort
- none · low · medium · high · xhigh
- Discovery tier
- Frontier
Pricing
USD per 1M tokens- Input
- $2.50
- Output
- $15.00
- Cache read
- $0.25
Reference prices as published by the model provider. API example
Specifications
- Context window
- 1.1M
- Max output
- 128K
- Average output (7 days)
- 409 tokens
- Longer than 41% of 465 models
Side by side
GPT 5.4 compared
Specifications, prices, speed and benchmark scores next to each other. Benchmarks below show every compared model too.
| GPT 5.4OpenAIThis model | DeepSeek V4 Flash 0423DeepSeekRemove | |
|---|---|---|
| Provider | OpenAI | DeepSeek |
| Discovery tier | Frontier | Open weights |
| Added | Mar 5, 2026 | Oct 7, 2026 |
| Limits | ||
| Context window | 1.05MBest | 1.05M |
| Max output | 128K | — |
| Average output | 409 tokens | 245 tokens |
| Pricing, USD per 1M tokens | ||
| Input | $2.50 | $0.14Best |
| Output | $15.00 | $0.28Best |
| Cache read | $0.25 | $0.028Best |
| Performance | ||
| Throughput | — | 121.3 tok/s |
| Time to first token | — | 2.4 s |
| Request success (24h) | 100.0%Best | 98.4% |
| Benchmark indexes | ||
| Intelligence Index | 18.2 | 18.9Best |
| Capabilities | ||
| Reasoning | Yes | Yes |
| Tools | Yes | Yes |
| Vision | Yes | No |
| Structured output | Yes | Yes |
| PDF input | Yes | No |
| Reasoning effort | none, low, medium, high, xhigh | none, low, high, max |
Public evaluations
Benchmarks
Independent evaluation results, with how each score ranks among the models in the Noumen catalog.
- GPT 5.4
- DeepSeek V4 Flash 0423
Intelligence Index
- GPT 5.418.2
- DeepSeek V4 Flash 042318.9
GPT 5.4: Better than 47% of 326 scored models
Document reasoning
AA-LCRv1.1
Reasoning across long documents
- GPT 5.4: 58.3%Top
- DeepSeek V4 Flash 0423: 41.7%
Reasoning and knowledge
Humanity’s Last Exam
Expert-level questions across academic fields
- GPT 5.4: 11.3%Top
- DeepSeek V4 Flash 0423: 7.8%
GPQA Diamond
Graduate-level science questions
- GPT 5.4: 74.8%Top
- DeepSeek V4 Flash 0423: 71.6%
IFBench
Following precise output instructions
- GPT 5.4: 48.4%Top
- DeepSeek V4 Flash 0423: 47.2%
Legacy evaluations
Terminal-Bench Hard
Earlier terminal task suite
- GPT 5.4: 37.9%Top
- DeepSeek V4 Flash 0423: 34.1%
τ²-Bench Telecom
Tool use in a customer-support simulation
- GPT 5.4: 36.0%
- DeepSeek V4 Flash 0423: 94.4%Top
AA-LCR (unversioned)
Long-context reasoning, earlier grading
- GPT 5.4: 58.3%Top
- DeepSeek V4 Flash 0423: 41.7%
Scores from Artificial Analysis, updated Oct 26, 2026. Artificial Analysis
- GPT 5.4
- DeepSeek V4 Flash 0423
| Leaderboard | GPT 5.4 | DeepSeek V4 Flash 0423 |
|---|---|---|
| Roleplay | #17 · 19.9% | Not in top 20 |
Ranks among Noumen catalog models on Use cases leaderboards. Only top-20 placements are listed.
- GPT 5.4
- DeepSeek V4 Flash 0423
| Leaderboard | GPT 5.4 | DeepSeek V4 Flash 0423 |
|---|---|---|
| Industry Mathematical | #19 · 1499.0 | Not in top 20 |
| Japanese | #14 · 1480.2 | Not in top 20 |
| Multi Turn | #14 · 1492.9 | Not in top 20 |
| Vision Style Control Overall | #16 · 1284.1 | Not in top 20 |
| Vision Style Control · Diagram | #11 · 1321.0 | Not in top 20 |
| Vision Style Control · English | #12 · 1280.9 | Not in top 20 |
| Vision Style Control · Homework | #10 · 1329.8 | Not in top 20 |
| Vision Style Control · Ocr | #16 · 1298.8 | Not in top 20 |
| Search Overall | #15 · 1199.1 | Not in top 20 |
| Document Overall | #11 · 1471.0 | Not in top 20 |
| Document Style Control Overall | #14 · 1470.1 | Not in top 20 |
Ranks among Noumen catalog models on LMArena leaderboards. Only top-20 placements are listed.LMArena
- GPT 5.4
- DeepSeek V4 Flash 0423
| Leaderboard | GPT 5.4 | DeepSeek V4 Flash 0423 |
|---|---|---|
| Epoch Capabilities Index | #14 · 159.6 | Not in top 20 |
| FrontierMath v2 (Tiers 1–3) | #2 · 82.5% | Not in top 20 |
| SWE-bench Verified | #4 · 76.9% | Not in top 20 |
| GPQA Diamond | #1 · 94.6% | Not in top 20 |
Ranks among Noumen catalog models on Epoch AI leaderboards. Only top-20 placements are listed.Epoch AI
- GPT 5.4
- DeepSeek V4 Flash 0423
| Leaderboard | GPT 5.4 | DeepSeek V4 Flash 0423 |
|---|---|---|
| ARC-AGI-1 | #18 · 94.5% | Not in top 20 |
| ARC-AGI-2 | #15 · 83.3% | Not in top 20 |
| ARC-AGI-3 (Default harness) | #17 · 0.2% | Not in top 20 |
Ranks among Noumen catalog models on ARC Prize leaderboards. Only top-20 placements are listed.ARC Prize
- GPT 5.4
- DeepSeek V4 Flash 0423
| Leaderboard | GPT 5.4 | DeepSeek V4 Flash 0423 |
|---|---|---|
| EQ-Bench 4 | #9 · 1260 | Not in top 20 |
| EQ-Bench 3 | #14 · 1537 | Not in top 20 |
Ranks among Noumen catalog models on EQ-Bench leaderboards. Only top-20 placements are listed.EQ-Bench
Live routing data
Performance and reliability
Recent averages across requests routed for this model. Your own speed depends on prompt length, reasoning effort and load.
- Prompt cache hits
- 79.7%
- About $0.71 per 1M input tokens with caching
- Request success (24h)
- 100.0%
- Reliable
Cache hits are the share of input tokens recently served from a prompt cache across all requests. They are not a forecast of your own hit rate.
- Healthy (98% or more)
- Degraded (90–98%)
- Down (under 90%)
- No requests
Developers
Use it through the API
Noumen’s API is OpenAI-compatible: any Chat Completions client works with a base URL and a key.
curl https://www.noumen.xyz/api/v1/chat/completions \
-H "Authorization: Bearer $NOUMEN_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "openai/gpt-5.4",
"reasoning_effort": "medium",
"messages": [{ "role": "user", "content": "Explain this stack trace." }]
}'import os
from openai import OpenAI
client = OpenAI(base_url="https://www.noumen.xyz/api/v1", api_key=os.environ["NOUMEN_API_KEY"])
reply = client.chat.completions.create(
model="openai/gpt-5.4",
messages=[{"role": "user", "content": "Explain this stack trace."}],
)
print(reply.choices[0].message.content)OpenAI
Related models
Other OpenAI models, starting with the same family.