AI models with video input: 53 that watch the footage

Video

Services found: 53

#	AI model	API $/1M (in / out)	Context window	Price / mo	Output $/1M	Overview
41	Qwen3.5 Plus 2026-04-20 Qwen	$0.3 / $1.8	1M tokens	API only	1.8	View
42	Seed-2.0-Lite Bytedance Seed	$0.25 / $2	262.1K tokens	API only	2	View
43	Seed 2.1 Turbo Bytedance Seed	$0.5 / $2.5	262.1K tokens	API only	2.5	View
44	GLM 4.6V Z.ai	$0.3 / $0.9	131.1K tokens	API only	0.9	View
45	Perceptron Mk1 Perceptron	$0.15 / $1.5	32.8K tokens	API only	1.5	View
46	Seed 1.6 Bytedance Seed	$0.25 / $2	262.1K tokens	API only	2	View
47	Seed-2.0-Code Bytedance Seed	$0.5 / $3	262.1K tokens	API only	3	View
48	Qwen3.8 Omni Flash Qwen		1M tokens	API only		View
49	MiMo-V2.6-Pro Xiaomi	$0.435 / $0.87	1M tokens	API only	0.87	View
50	MiMo-V2.6-Flash Xiaomi	$0.14 / $0.28	1M tokens	API only	0.28	View
51	MiMo-V2.6-Pro-UltraSpeed Xiaomi	$4.35 / $8.7	1M tokens	API only	8.7	View
52	Perceptron Mk1.5 Perceptron	$0.15 / $1.5	36.9K tokens	API only	1.5	View
53	Ling 3.0 Flash VL InclusionAI	$0.02 / $0.059	256K tokens	API only	0.059	View

Browse by need

Reasoning models169 models

Coding assistants17 models

Vision and images156 models

Long context 1M+82 models

Free tier available80 models

Open weights88 models

Fifty-three models accept video as input - describing what happens on screen, summarising a recording, answering questions about a clip. This is comprehension, not generation.

Kimi K3, GLM 5.3 Flash and Gemini 3.8 Flash lead, with output from $0.25 to $15 per million tokens. Video is billed by sampled frames plus audio, so a minute of footage can consume more tokens than a long document and costs climb faster than intuition suggests. Sampling rate is usually configurable; dropping it is the first optimisation.

Frames, not minutesBilling counts sampled frames plus audio, so cost scales with sampling rate as much as length.

Comprehension onlyThese models watch and explain; none of them produce footage.

Sampling is the leverHalving the frame rate usually halves the bill with little loss on slow-moving content.

Models with this capability

The 53 models below all support this capability. Prices are input and output per 1M tokens from the vendor official price list; for dirhams multiply by 3.6725. Tap a card to open the model page.

GLM 5.3 FlashX

Z.ai

in $0.37out $1.251M tokens

Kimi K3

Moonshot AI

in $3out $151.05M tokens

GLM 5.3 Flash

Z.ai

in $0.075out $0.251.31M tokens

Gemini 3.8 Flash

Google DeepMind

in $0.75out $3.751.05M tokens

How to choose

Set sampling to match the content — a lecture needs far fewer frames than a sports clip.
Check maximum clip length and whether the model accepts audio alongside video.
Pre-trim footage: sending only the relevant minutes beats sending the whole recording.
Verify timestamp accuracy if you rely on the model to locate moments.

Common mistakes

Uploading raw recordingsAn hour of meeting footage costs more than the transcript of the same hour by a wide margin.

Expecting frame-perfect recallSampling means the model literally does not see every frame.

Who this suits

OperationsReviewing recorded processes for compliance or quality.

EducationTurning lectures into structured notes and chapters.

Media teamsLogging and tagging archives without watching them.

Related filters

Vision156 models that read images

Audio understanding25 models that interpret speech and sound

Realtime voice7 models for live voice conversation

Long context 1M+82 models holding 1M tokens or more

All 239 modelsThe full rating with filters and prices

Frequently asked questions

Is this video generation?

No, this is comprehension. These models watch and describe; they do not produce footage.

Why is video so expensive?

Billing counts sampled frames plus audio, so a minute of footage can exceed a long document. Lowering the sampling rate is the first optimisation.

What is it good for?

Meeting and lecture summaries, surveillance triage, quality control on recorded processes, accessibility descriptions.

Can it read text on screen?

Yes, subject to resolution — screen recordings work better than camera footage of screens.

Does it hear the audio track?

On most models yes, and it materially improves summaries.

Ratings 1

Avg. score 5

Rate this page

Help us improve