← Back to directory
OpenAI
GPT-4.1
Released Apr 14, 2025 · knowledge cutoff 2024-06
- Status
- Superseded
- Location
- United States
- Modality
- Multimodal
- Context window
- 1M
- Max output
- 33K
- Speed
- —
- Price ($/MTok in / out)
- $2 / $8
- Cost per task
- —
- Open weights
- No
Benchmarks6 of 8 reported
MMLU-Pro
80.5%
GPQA Diamond
66.3%
SWE-bench Verified
54.6%
AIME (latest)
48.1%
Humanity's Last Exam
5.4%
LMArena Elo
1366
Terminal-Bench 2.1
—
ARC-AGI-2
—
Strengths & weaknesses
Strengths
- Instruction adherence is the whole point — it honours format rules and negative constraints literally
- Predictable latency and behavior, which made it the safe choice for high-volume pipelines
- Emits targeted diffs rather than rewriting whole files
Weaknesses
- Literal to a fault: needs spelling out where a reasoning model would infer intent
- No thinking mode, so genuinely hard problems stay out of reach at any prompt length
- Long-context recall degrades well before the window's nominal limit
Workhorse non-reasoning line before GPT-5 unified the stack.
For developers
API model strings
Not researched
Licence
Proprietary — weights not released
Retirement
No retirement announced
Lineage
No recorded predecessor