Is YOLO12 better than YOLOv8?

On published COCO accuracy, YOLO12. It leads 5 of the 5 size tiers where both families publish a figure, by an average of 2.2 mAP. The gap is widest at the nano tier, where YOLO12 is ahead by 3.3 mAP. YOLO12 tops out at 55.2 mAP and YOLOv8 at 53.9. "Better" still depends on your constraint, though: accuracy is one axis and parameter count, task coverage and licence are the others, all of which are in the table above.

Which is smaller, YOLO12 or YOLOv8?

YOLO12 gets more out of each parameter. YOLO12l reaches 53.7 mAP on 26.4M parameters, matching or beating YOLOv8l at 43.7M: about 40% fewer parameters for the same accuracy. That is the comparison worth making, since it is the same result from a smaller checkpoint.

Is YOLO12 faster than YOLOv8?

Nothing on this page is a speed measurement, and it would be dishonest to imply one: every figure here is published COCO accuracy and parameter count. Parameters are a rough guide to inference cost within one family, but they do not transfer across architectures, and the post-processing step matters as much as the backbone. Export both to ONNX and measure on your own hardware and footage, which is the only comparison that predicts what you will actually get.

Are YOLO12 and YOLOv8 free for commercial use?

Both ship under AGPL-3.0, so the licence is not a tiebreaker here: choose on accuracy and size. AGPL-3.0 is free for open-source, research and internal use, but a closed-source product or paid service needs an Ultralytics Enterprise licence. Read the licence text before you ship; this is a plain reading of it, not legal advice.

Model comparison · YOLO12 vs YOLOv8

YOLO12 vs YOLOv8

A size-tier-by-size-tier comparison of YOLO12 and Ultralytics YOLOv8: published COCO accuracy, parameter counts, licensing and which one to reach for. Every figure is the model authors' own, linked to its source.

CompareYOLO12

AgainstYOLOv8

40%fewer params

mAP comparison

YOLO12YOLOv8

YOLO12

Nano 40.6
Small 48.0
Medium 52.5
Large 53.7
Extra large 55.2

YOLOv8

Nano 37.3
Small 44.9
Medium 50.2
Large 52.9
Extra large 53.9

COCO mAP at 640px, as published by each model’s authors. The y axis spans 34.5 to 58.0 rather than starting at zero, so small differences stay visible. Points are the published checkpoints; the joining lines are for reading order only.

Accuracy and size, tier by tier

Each row pairs the equivalent rung of both size ladders, so the numbers sit side by side at comparable capacity. Every figure is the model authors’ own.

COCO mAP50-95 @ 640px
mAP: higher is better
Params (M): lower is better

YOLO12 and YOLOv8 compared at each size tier: published COCO mAP50-95 at 640px and parameter counts in millions.
Tier	YOLO12			YOLOv8			Δ mAP
Tier	Model	mAP	Params	Model	mAP	Params	Δ mAP
Nano	YOLO12n	40.6	2.6M	YOLOv8n	37.3	3.2M	+3.3
Small	YOLO12s	48.0	9.3M	YOLOv8s	44.9	11.2M	+3.1
Medium	YOLO12m	52.5	20.2M	YOLOv8m	50.2	25.9M	+2.3
Large	YOLO12l	53.7	26.4M	YOLOv8l	52.9	43.7M	+0.8
Extra large	YOLO12x	55.2	59.1M	YOLOv8x	53.9	68.2M	+1.3

Δ mAP is YOLO12 minus YOLOv8: positive means YOLO12 is ahead at that tier. Sources: YOLO12 docs · YOLO12 paper · YOLOv8 docs

Additional information

YOLO12 posts the higher published COCO mAP at all 5 matched size tiers, averaging 2.16 mAP ahead of YOLOv8. The gap is widest at the nano tier, where YOLO12 leads by 3.3 mAP. At the top of each range, YOLO12 reaches 55.2 mAP against 53.9.
40% fewer parameters, same accuracy. YOLO12l reaches 53.7 mAP from 26.4M parameters, matching or beating YOLOv8l at 52.9 mAP from 43.7M.
Parameter count is not latency. It is a good proxy for download size and memory, but throughput depends on your hardware, batch size and export format, so measure on your own target.

YOLO12

The attention-centric generation, between YOLO11 and YOLO26.

Released: 2025 · Tian, Ye & Doermann (UB / UCAS)
License: AGPL-3.0
Framework: Ultralytics (PyTorch)
Tasks: Detect, Segment, Classify, Pose, OBB

An attention-centric YOLO: area attention and residual ELAN blocks replace some of the pure-convolution stack, buying about a point of mAP over YOLO11 at a similar size.

Download YOLO12 weights →

YOLOv8

The workhorse that made anchor-free multi-task YOLO mainstream.

Released: 2023 · Ultralytics
License: AGPL-3.0
Framework: Ultralytics (PyTorch)
Tasks: Detect, Segment, Classify, Pose, OBB

The anchor-free model that made multi-task YOLO mainstream: one architecture for detection, segmentation, pose, classification and OBB, with the largest ecosystem.

Download YOLOv8 weights →

YOLO12 and YOLOv8 against the whole lineage

YOLO12 and YOLOv8 against every other detector here, on one shared scale. The rest are dimmed; hover any row to bring it back.

Every YOLO detector in this section, with release year, publisher, published COCO mAP50-95 range at 640px, parameter-count range and licence.
Model	Released	mAP range	Params range	Licence
YOLO26	2026	40.9 to 57.5	2.4M to 55.7M	AGPL-3.0
YOLO12	2025	40.6 to 55.2	2.6M to 59.1M	AGPL-3.0
YOLO11	2024	39.5 to 54.7	2.6M to 56.9M	AGPL-3.0
YOLOv10	2024	38.5 to 54.4	2.3M to 29.5M	AGPL-3.0
YOLOv9	2024	38.3 to 55.6	2.0M to 58.1M	GPL-3.0
YOLOv8	2023	37.3 to 53.9	3.2M to 68.2M	AGPL-3.0
YOLOv5 (v5u)	2020	34.3 to 53.2	2.6M to 97.2M	AGPL-3.0

Frequently asked questions

Is YOLO12 better than YOLOv8?: On published COCO accuracy, YOLO12. It leads 5 of the 5 size tiers where both families publish a figure, by an average of 2.2 mAP. The gap is widest at the nano tier, where YOLO12 is ahead by 3.3 mAP. YOLO12 tops out at 55.2 mAP and YOLOv8 at 53.9. "Better" still depends on your constraint, though: accuracy is one axis and parameter count, task coverage and licence are the others, all of which are in the table above.
What is the difference between YOLO12 and YOLOv8?: YOLO12 (Tian, Ye & Doermann (UB / UCAS), 2025) and YOLOv8 (Ultralytics, 2023) are 2 years apart. The attention-centric generation, between YOLO11 and YOLO26. The workhorse that made anchor-free multi-task YOLO mainstream. Both cover 5 tasks.
Which is smaller, YOLO12 or YOLOv8?: YOLO12 gets more out of each parameter. YOLO12l reaches 53.7 mAP on 26.4M parameters, matching or beating YOLOv8l at 43.7M: about 40% fewer parameters for the same accuracy. That is the comparison worth making, since it is the same result from a smaller checkpoint.
Is YOLO12 faster than YOLOv8?: Nothing on this page is a speed measurement, and it would be dishonest to imply one: every figure here is published COCO accuracy and parameter count. Parameters are a rough guide to inference cost within one family, but they do not transfer across architectures, and the post-processing step matters as much as the backbone. Export both to ONNX and measure on your own hardware and footage, which is the only comparison that predicts what you will actually get.
Are YOLO12 and YOLOv8 free for commercial use?: Both ship under AGPL-3.0, so the licence is not a tiebreaker here: choose on accuracy and size. AGPL-3.0 is free for open-source, research and internal use, but a closed-source product or paid service needs an Ultralytics Enterprise licence. Read the licence text before you ship; this is a plain reading of it, not legal advice.

Every matchup

Looking for the weights themselves? Download any of these checkpoints from the model-weights hub, or see how the trackers behind them compare in the YOLO tracker benchmarks.