OpenAI · released Mar 5, 2026
GPT-5.4
GPT-5.4 is an older OpenAI model that the GPT-6 family has passed.
8 months oldThinks before answering
- Input price
- $2.5 / M
- Output price
- $15 / M
- Context
- 1.1M tokens
- Longest reply
- 128K
- Reads
- text, image, pdf
- Knows up to
- 2025-08-31
Where it shines
- It scores 1303 on vision.
- It scores 81.8 on Terminal-Bench.
Where it struggles
- Its web dev score of 1401 is among the lowest here.
Where you can use it
- OpenAI APIAPI$2.50 in / $15 out per million tokens.
- OpenRouterAPIListed as
openai/gpt-5.4
How it scores on each job
Scores run 0 to 100 and blend every measure we have for that job. The number after the dot is its rank among the 33 models in the lineup.
The evidence
| Measure | Score | Rank in lineup | Source |
|---|---|---|---|
| Epoch Capabilities Index | 157 | #15 | Epoch AI |
| LMArena text | 1469 | #16 | LMArena |
| LMArena hard prompts | 1485 | #15 | LMArena |
| LMArena instruction following | 1469 | #16 | LMArena |
| LMArena multi-turn | 1482 | #11 | LMArena |
| LMArena creative writing | 1439 | #17 | LMArena |
| LMArena writing & language | 1458 | #15 | LMArena |
| LMArena coding | 1497 | #15 | LMArena |
| LMArena math | 1488 | #14 | LMArena |
| LMArena business & finance | 1475 | #10 | LMArena |
| LMArena legal & government | 1486 | #11 | LMArena |
| LMArena medicine | 1458 | #18 | LMArena |
| WebDev Arena | 1401 | #28 | LMArena |
| Vision Arena | 1303 | #7 | LMArena |
| Search Arena | 1199 | #7 | LMArena |
| Document Arena | 1471 | #4 | LMArena |
| SWE-bench Verified | 77% | #3 | Epoch AI |
| Terminal-Bench | 82% | #2 | Epoch AI |
| GPQA Diamond | 93% | #12 | Epoch AI |
| Humanity's Last Exam | 36% | #6 | Epoch AI |
| FrontierMath | 79% | #11 | Epoch AI |
| SimpleQA Verified | 45% | #17 | Epoch AI |
| METR time horizon | 5.7 h | #2 | Epoch AI |
| ARC-AGI-2 | 74% | #12 | Epoch AI |
| DeepResearch Bench | 35 | #3 | Epoch AI |
Compare it with these
4 months old
OpenAI
GPT-5.6 Terra
GPT-5.6 Terra is OpenAI's older mid-tier model, and GPT-6.1 Terra now costs half as much.
Long documents
- Price
- $2 in · $12 out
- Context
- 1.1M
6 months old
OpenAI
GPT-5.5
GPT-5.5 still codes and reads documents well, and newer OpenAI models now cost less.
CodingLong documents
- Price
- $5 in · $30 out
- Context
- 1.1M
Open weights4 months old
Moonshot
Kimi K3
Kimi K3 is Moonshot's July open-weight model, and Kimi K3.1 Pro has replaced it as the newest Kimi.
Professional workEveryday questions
- Price
- $3 in · $15 out
- Context
- 1M