GPT-4.1

Name: GPT-4.1
Author: OpenAI

OpenAI · 2025-04-14

GPT-4.1 is OpenAI's GPT-4.1 family flagship, a non-reasoning model tuned especially for coding and instruction-following with a long (up to ~1M token) context window. It improves on GPT-4o for software-engineering and agentic workflows. Its only published snapshot, and the gpt-4.1 alias's default, is gpt-4.1-2025-04-14.

Benchmark scores

Benchmark	Score
A-Fantasia	53.0
A-Fantasia (Valid Attempts with Retries)	53.0
A-Fantasia Backwards Spelling	64.0
A-Fantasia Chess	13.0
A-Fantasia Chess (Valid Attempts with Retries)	13.0
A-Fantasia Cube (Valid Attempts with Retries)	82.0
A-Fantasia Cube Rotation	82.0
A-Fantasia Spelling (Valid Attempts with Retries)	64.0
AA-LCR	61.0
AA-LCR v1.1	68.3
AA-Omniscience Index	-39.6
AidanBench	1356
Aider Polyglot	52.4
AIME 2024 (avg@8)	48.8
AIME 2024-2025	39.6
AIME 2025 (Artificial Analysis avg@10)	34.7
AIME 2025 (avg@8)	30.4
ALE-Bench	985
ARC-AGI-1 Semi-Private	5.5
ARC-AGI-2	0.4
Arena Text — Business, Management & Financial Ops (No Style Control)	1369
Arena Text — Chinese (No Style Control)	1382
Arena Text — Creative Writing (No Style Control)	1363
Arena Text — English (No Style Control)	1392
Arena Text — Entertainment, Sports & Media (No Style Control)	1365
Arena Text — Exclude Ties (No Style Control)	1356
Arena Text — Exclude Ties (Style Controlled)	1403
Arena Text — Expert (No Style Control)	1364
Arena Text — French (No Style Control)	1382
Arena Text — French (Style Controlled)	1420