HeyGen Video tops 8 AI video models on cost per usable second at 8.6/10, priced on 1,000 seconds a month. One costs 13x the winner and scores 7.4.
Per-second prices for AI video now run from 3 cents to 40 cents, and the headline number tells you less than it seems. A cheap rate often applies only to 480p, a "1080p" rate can quietly drop the audio, and a short clip cap means you pay for seconds you throw away when you stitch shots together.
HeyGen Video came out on top at 8.6 out of 10 on AI video cost per second, at $0.03 a second for 768p clips with native sound once its October promotion ends.
I priced eight models against one job, 1,000 generated seconds a month at each model's cheapest tier of 720p or above, and scored each on the same evidence checklist. LTX-2.3 finished a hair behind, and it wins outright if your delivery target is 1080p.
The short answer
- Best overall: HeyGen Video, 8.6/10. $0.03 a second at 768p with sound, $360 a year for 1,000 seconds a month (standard rate from November 2026).
- Best for 1080p and 4K: LTX-2.3, 8.5/10. $0.04 a second at 1080p on its Fast tier, $480 a year for the same job, with open weights.
- Best Google option: Veo 3.1 Fast, 8.2/10. $0.10 a second at 720p with audio included, $1,200 a year.
- Cheapest absolute rate: HeyGen Video at $0.015 a second at 768p through October 31, 2026, during its launch promotion.
- Overpriced for this job: Veo 3.1 Standard costs 13.3x the winner's 12-month total and scores 7.4.
- Method: one priced job, one evidence checklist, eight models, six weighted categories, prices dated July to October 2026.
The Short Version
- HeyGen Video, 8.6 Best overall. Lowest standard per-second rate here with native sound and fixed seeds. For teams generating B-roll and product shots at volume.
- LTX-2.3, 8.5 Best for 1080p and 4K. $0.04 a second at 1080p and open weights. For teams whose delivery target is full HD or higher.
- Veo 3.1 Fast, 8.2 Best Google option. Audio included in the rate and a Lite tier below it. For teams already on Google Cloud.
- Seedance 2.0 Fast, 7.9 Best 1080p value among closed models. About $0.09 a second through API resellers. For social teams that need full HD without upscaling.
- Runway Gen-4.5, 7.8 Best creative controls. $0.12 a second through the API, with an editor around it. For editors who want keyframes and a timeline.
- Kling 3.0, 7.7 Best for longer single shots. Around $0.13 a second through API resellers. For narrative clips that need motion over 10 seconds.
- Wan 2.6, 7.5 Best open model for drafts. About $0.07 to $0.10 a second at 720p hosted. For teams that will self-host later.
- Veo 3.1, 7.4 Best top-end quality and 4K. $0.40 a second with audio. For hero shots where the per-second rate does not matter.
Prices as published or captured in July to October 2026; each entry carries its own date.
Why I Judge Models This Way
I have bought generated footage for product and marketing teams since per-second billing became the norm, and the bill is always set by the hundredth clip, not the first. The question I judge every model against is the one a producer asks before a monthly budget review: what does a usable second of footage with sound cost me at the resolution I deliver?
A month of stock B-roll licensing is the floor most teams compare against, and per-second pricing only wins if the rate holds at the resolution and length you need.
How I Compared
I priced every model against the same job, 1,000 generated seconds a month at its cheapest tier of 720p or above with sound where the model offers it, and checked the same evidence for every model: the vendor's published API rates, resolution and clip-length limits, audio rules, and dated third-party pricing captures from July to October 2026.
No score depends on a render I did not see documented, and I did not add a retry multiplier, so heavy retry workflows should scale every row by the same factor.
Cost per usable second (30%)
I scored the published per-second rate at the job's tier, including whether audio costs extra, and the 12-month total for 12,000 seconds.
Resolution and clip length (22%)
I scored the highest resolution available, the price multiplier to reach 1080p, and the maximum clip length, since short caps force stitching.
Output quality (18%)
I scored documented output quality from vendor materials and independent arena coverage, with vendor-run benchmarks labeled as such.
Audio and control (12%)
I scored native audio, fixed seeds, image and reference conditioning, and camera or keyframe controls.
Access and billing (9%)
I scored how the model is sold: first-party API, resellers, subscription credits, and minimum spend.
Licensing and exit (9%)
I scored open weights, commercial-use terms, and whether you can leave without losing the ability to regenerate.
Rubric anchors: 9 to 10 means the model fully covers the category and does something no other model here does. 7 to 8 means it covers the category with one documented limitation. 5 to 6 means it needs a workaround. 3 to 4 means a documented limit or price changes the purchase. The weights sum to 100.
The Two Things That Decided This
The first was the resolution multiplier. Most headline rates apply to the lowest tier a model sells. Veo 3.1 Fast moves from $0.10 to $0.12 a second between 720p and 1080p on Google's Gemini API, LTX-2.3 Fast moves from $0.04 at 1080p to $0.16 at 4K on fal, and HeyGen's developer docs bill 1080p and 2K at 3x the 768p rate. The cheapest second is only cheap at the resolution it is quoted for.
The second was the clip cap. A model that stops at 5 or 8 seconds turns a 10-second shot into two generations and a cut. Atlas Cloud's comparison notes that Wan 2.6 tops out at 5 seconds, and Veo 3.1 runs 8-second clips at 1080p, while HeyGen Video runs 5 to 15 seconds and LTX-2.5 runs up to 20.
Three Kinds of Model That Are Not Competing With Each Other
Some models here are sold for different jobs, so I ranked them inside tiers. Numbering runs 1 to 8 across the article.
Tier 1: Volume models. Low per-second rates for B-roll, product shots, and social clips at 768p to 1080p. This tier is not where you buy a 4K hero shot.
Tier 2: Premium closed models. Higher rates for stronger motion, longer shots, or creative controls. This tier cannot hit volume budgets at 1,000 seconds a month without a large line item.
Tier 3: Open and top-end models. Open weights you can self-host, and the most expensive closed model for hero shots. This tier asks for either engineering time or a big budget.
Tier 1: Volume Models
1. HeyGen Video: Best Overall on Cost per Second
- Score: 8.6/10
- Entry price: $0.03 per second at 768p (standard rate from November 2026); $0.015 through October 31, 2026
- Cost for 1,000 seconds a month: $360/year at the standard rate
- Metering: Per second of output; 480p $0.02, 768p $0.03, reference mode $0.06 at standard rates
- Volume cliff: None published; prepaid API wallet from $5
- Free tier: No free API credits
- Priced on: HeyGen API standard rates, October 2026 help center
HeyGen Video clears the job at $360 a year, the lowest 12-month total here.
What it wins: top-tier visuals at a fraction of the usual per-second price. HeyGen Video generates a whole scene in one call, subject, setting, light, and a matching track of dialogue, ambience, and sound effects, at 3 cents a second at 768p.
How it works: you describe the shot in text, or start from an image used as the literal first frame, or pass up to 12 reference images, videos, and audio clips to hold a product's colour and markings. Fixed seeds return the same clip for the same prompt, so a shot can be refined one clause at a time.
Where it falls short: resolution and length. Clips cap at 15 seconds, and 1080p and 2K bill at 3x the 768p rate, which is where LTX-2.3 wins.
Client example: No independent client result with a time or cost delta has been published since the September 30, 2026 launch; HeyGen reports a first-place image-to-video ranking in its own testing.
Pros:
- $0.03 a second at 768p with native sound at the standard rate
- Reference mode holds a product's colour, form, and markings from up to 12 references
- Fixed seeds reproduce a shot for one-change-at-a-time iteration
- Prompts up to 32,000 characters for precise direction
Cons:
- Clips run 5 to 15 seconds, so longer shots are assembled
- Detailed prompts produce materially better output than short ones
Category scores: Cost 9.5 | Resolution and length 7.4 | Quality 8.8 | Audio and control 9.0 | Access 8.0 | Licensing 8.4
2. LTX-2.3: Best for 1080p and 4K
- Score: 8.5/10
- Entry price: $0.04 per second at 1080p (Fast tier, fal, July 2026)
- Cost for 1,000 seconds a month: $480/year
- Metering: Per second; Pro $0.06 at 1080p, Fast $0.16 at 4K
- Volume cliff: None published
- Free tier: Open weights and a free desktop app
- Priced on: fal LTX-2.3 Fast, July 2026 capture
LTX-2.3 clears the job at $480 a year at 1080p, a third more than HeyGen Video but at a higher resolution.
What it wins: price at full HD and above. LTX-2.3's lowest hosted tier starts at 1080p, so there is no cheaper low-resolution rate to be caught out by.
How it works: Lightricks publishes first-party per-second API rates, and the model also ships as open weights, so a team can prototype on the hosted API and move to its own GPUs later. LTX-2.5, released in August 2026, raised the Fast rate to $0.09 at 720p and $0.13 at 1080p on the official API.
Where it falls short: output quality trails the premium closed models in independent coverage, and I could not confirm audio rules on every hosted tier.
Client example: No verifiable public client result with a time or cost delta was found.
Pros:
- $0.04 a second at 1080p on the Fast tier
- 4K available at $0.16 a second
- Open weights for self-hosting
- First-party published per-second pricing
Cons:
- LTX-2.5 costs roughly 2x to 3x more per second than LTX-2.3 at shared resolutions
- Audio availability varies by tier and host
Category scores: Cost 9.2 | Resolution and length 9.0 | Quality 7.6 | Audio and control 7.0 | Access 8.6 | Licensing 9.0
3. Veo 3.1 Fast: Best Google Option
- Score: 8.2/10
- Entry price: $0.10 per second at 720p, $0.12 at 1080p (Gemini API, August 2026)
- Cost for 1,000 seconds a month: $1,200/year at 720p
- Metering: Per second; audio included; 4K $0.30
- Volume cliff: None published
- Free tier: None for video generation
- Priced on: Google Gemini API, August 2026 capture
Veo 3.1 Fast clears the job at $1,200 a year at 720p.
What it wins: audio and ecosystem. Google includes audio in every Veo 3.1 rate, and the model sits inside Gemini API and Vertex billing teams already have.
How it works: the Veo 3.1 family has three tiers: Lite from $0.05 a second at 720p, Fast from $0.10, and Standard at $0.40. Fast is the balance point for teams that want Veo motion without the Standard rate.
Where it falls short: cost is 3.3x HeyGen Video's for the same job, and 1080p clips run 8 seconds.
Client example: No verifiable public client result with a time or cost delta was found.
Pros:
- Audio included in the per-second rate
- Three tiers from $0.05 to $0.40 a second
- 4K available at $0.30 a second
- Inside existing Google Cloud billing
Cons:
- $1,200 a year for the job, 3.3x the top pick
- 1080p clips cap at 8 seconds
Category scores: Cost 7.6 | Resolution and length 8.4 | Quality 8.6 | Audio and control 8.8 | Access 8.8 | Licensing 8.0
Tier 2: Premium Closed Models
4. Seedance 2.0 Fast: Best 1080p Value Among Closed Models
- Score: 7.9/10
- Entry price: About $0.09 per second at 1080p (Atlas Cloud, 2026)
- Cost for 1,000 seconds a month: $1,080/year
- Metering: Per second through API resellers
- Volume cliff: None published
- Free tier: None found
- Priced on: Seedance 2.0 Fast via Atlas Cloud, 2026 capture
Seedance 2.0 Fast clears the job at about $1,080 a year at 1080p.
What it wins: full HD from a closed model at under 10 cents a second, with output Atlas Cloud describes as often usable without post-processing.
How it works: Seedance 2.0 is ByteDance's model, sold mostly through resellers. The standard tier sits far higher: HeyGen's published comparison chart lists Seedance 2.0 at $0.303 a second at 768p.
Where it falls short: access runs through resellers whose rates vary, and the standard tier costs over 3x the Fast tier.
Client example: No verifiable public client result with a time or cost delta was found.
Pros:
- About $0.09 a second at 1080p on the Fast tier
- Output often usable without upscaling, per Atlas Cloud
- Strong camera choreography in independent coverage
- Available through several API hosts
Cons:
- Rates vary by reseller, so budgets need a fixed host
- The standard tier lists at about $0.30 a second
Category scores: Cost 7.8 | Resolution and length 8.0 | Quality 8.4 | Audio and control 8.4 | Access 7.2 | Licensing 7.0
5. Runway Gen-4.5: Best Creative Controls
- Score: 7.8/10
- Entry price: $0.12 per second via the API (12 credits at $0.01); Standard plan $12/mo billed annually
- Cost for 1,000 seconds a month: $1,440/year via the API
- Metering: Credits; Gen-4.5 12 per second, Gen-4 Turbo 5
- Volume cliff: Standard's 625 monthly credits cover about 52 seconds of Gen-4.5
- Free tier: 125 one-time credits
- Priced on: Runway API and plans, October 2026
Runway Gen-4.5 clears the job at $1,440 a year through the API.
What it wins: the editor. Keyframes, camera controls, and a timeline sit around the model, which no other entry here offers.
How it works: Gen-4.5 fell from 25 to 12 credits a second in June 2026, per Flexprice's change log. Gen-4 Turbo at 5 credits a second makes drafts cheap before finals on Gen-4.5.
Where it falls short: cost against the volume tier, and Standard and Pro subscription credits do not roll over.
Client example: No verifiable public client result with a time or cost delta was found.
Pros:
- Keyframes, camera controls, and an editor around the model
- Gen-4 Turbo drafts at 5 credits a second
- API credits at $0.01 each
- Partner models such as Veo 3.1 available in the same account
Cons:
- $1,440 a year for the job, 4x the top pick
- Subscription credits on Standard and Pro do not roll over
Category scores: Cost 7.2 | Resolution and length 7.6 | Quality 8.8 | Audio and control 7.6 | Access 8.6 | Licensing 7.8
6. Kling 3.0: Best for Longer Single Shots
- Score: 7.7/10
- Entry price: About $0.13 per second ($1.26 per 10-second clip, Atlas Cloud, 2026)
- Cost for 1,000 seconds a month: About $1,512/year
- Metering: Per second through resellers; subscription credits in Kling's own app
- Volume cliff: None published
- Free tier: Limited credits in the consumer app
- Priced on: Kling 3.0 via Atlas Cloud, 2026 capture
Kling 3.0 clears the job at about $1,512 a year.
What it wins: 10-second single shots. Atlas Cloud lists Kling 3.0 among the few models that generate 10-second clips in one call.
How it works: Kling sells subscription credits in its own app and per-second rates through resellers. HeyGen's comparison chart lists Kling 3.0 Pro at $0.168 a second at 768p, above the reseller rate.
Where it falls short: cost, and rates differ between Kling's app, its Pro tier, and resellers.
Client example: No verifiable public client result with a time or cost delta was found.
Pros:
- 10-second clips in one generation
- Photorealistic humans and complex motion in independent coverage
- Multi-shot storyboard mode
- Several hosting routes
Cons:
- About $1,512 a year for the job, 4.2x the top pick
- Pricing differs across the app, Pro tier, and resellers
Category scores: Cost 7.0 | Resolution and length 8.2 | Quality 8.6 | Audio and control 8.2 | Access 7.4 | Licensing 7.0
Tier 3: Open and Top-End Models
7. Wan 2.6: Best Open Model for Drafts
- Score: 7.5/10
- Entry price: About $0.07 per second at 720p (Atlas Cloud); $0.10 at 720p and $0.15 at 1080p (ElevenLabs scale-tier reference)
- Cost for 1,000 seconds a month: About $840/year at $0.07
- Metering: Per second through hosts
- Volume cliff: 5-second clip cap per Atlas Cloud
- Free tier: Open weights
- Priced on: Wan 2.6 via Atlas Cloud, 2026 capture
Wan 2.6 clears the job at about $840 a year hosted, or for GPU time if self-hosted.
What it wins: an open model that is cheap to draft on and free to run on your own hardware.
How it works: Wan is Alibaba's open model family, hosted by many API providers at different rates. The 5-second cap means a 10-second shot needs two generations.
Where it falls short: clip length and resolution, which is why it scored lowest on that category.
Client example: No verifiable public client result with a time or cost delta was found.
Pros:
- Open weights for self-hosting
- About $0.07 a second at 720p on Atlas Cloud
- Many hosting providers
- Unrestricted prompts on self-hosted runs
Cons:
- 5-second clips force stitching for longer shots
- Hosted rates vary from $0.07 to $0.15 a second
Category scores: Cost 8.2 | Resolution and length 6.0 | Quality 7.2 | Audio and control 7.0 | Access 8.0 | Licensing 9.0
8. Veo 3.1: Best Top-End Quality and 4K
- Score: 7.4/10
- Entry price: $0.40 per second at 720p to 1080p, $0.60 at 4K (Gemini API, August 2026)
- Cost for 1,000 seconds a month: $4,800/year
- Metering: Per second; audio included
- Volume cliff: None published
- Free tier: None for video generation
- Priced on: Google Gemini API, August 2026 capture
Veo 3.1 Standard clears the job at $4,800 a year.
What it wins: quality at the top end, with 4K and audio included in the rate.
How it works: the same Gemini billing as Veo 3.1 Fast, at 4x the rate. fal sells Veo 3.1 at $0.20 a second with audio off and $0.40 with audio on, so turning sound off halves the bill there.
Where it falls short: cost, which is why it is the overpriced pick for a volume job.
Client example: No verifiable public client result with a time or cost delta was found.
Pros:
- Highest documented quality tier in the Veo family
- 4K at $0.60 a second
- Audio included in Google's rate
- Strong prompt adherence in independent coverage
Cons:
- $4,800 a year for the job, 13.3x the top pick
- 1080p clips cap at 8 seconds
Category scores: Cost 4.0 | Resolution and length 8.8 | Quality 9.4 | Audio and control 9.0 | Access 8.8 | Licensing 8.0
What 1,000 Seconds a Month Costs
The job: 1,000 generated seconds a month at each model's cheapest tier of 720p or above, with sound where the model includes it. That is 12,000 seconds a year, about a hundred 10-second clips a month.
Prices from vendor pages and dated third-party captures, July to October 2026.
How to check one row: HeyGen Video's standard rate is $0.03 a second at 768p, so 1,000 seconds a month costs $30 and 12,000 seconds a year cost $360. Through October 31, 2026, the promotional $0.015 rate halves that.
Resolution changes the ranking. At 1080p, HeyGen Video bills at 3x its 768p rate, about $0.09 a second, and LTX-2.3 Fast at $0.04 becomes the cheapest model in the table.
The crossover. HeyGen Video is cheapest for any volume at 768p. If every clip must ship at 1080p, LTX-2.3 is cheapest; if every clip needs Google billing and audio included, Veo 3.1 Lite at $0.05 a second at 720p is the lowest Veo rate.
Which model is overpriced for this job. Veo 3.1 Standard costs $4,800 a year against HeyGen Video's $360, which is 13.3x, and it scored 7.4 against 8.6. What the extra $4,440 buys is the top-end Veo quality and 4K, which matters for hero shots rather than a hundred clips a month.
AI Video Cost per Second Compared: Specs
Verified against vendor pages and dated captures, July to October 2026.
Which One Is For You
By the work you are doing
- If you generate B-roll and product shots at volume, use HeyGen Video.
- If every clip ships at 1080p or 4K, use LTX-2.3.
- If your team works inside Google Cloud, use Veo 3.1 Fast.
- If you edit with keyframes and a timeline, use Runway Gen-4.5.
By budget
- Free: LTX-2.3 or Wan 2.6 open weights on your own hardware.
- Under $50 a month: HeyGen Video at $30 for 1,000 seconds at 768p.
- Over $300 a month: Veo 3.1, for hero shots where quality outranks rate.
By learning-curve tolerance
- I want output today: HeyGen Video through OpenRouter or the HeyGen API with a $5 wallet.
- I will spend a weekend: Runway Gen-4.5, where the editor rewards practice.
- I have a team to train: LTX-2.3, where self-hosting pays back at scale.
Too New To Rank
LTX-2.5 launched on August 11, 2026 with Fast at $0.09 a second at 720p and $0.13 at 1080p on the official API. It enters the ranking once independent quality coverage separates it from LTX-2.3. I will look again in January 2027.
Veo 3.1 Lite starts at $0.05 a second at 720p on the Gemini API. It would enter Tier 1 once its quality against Veo 3.1 Fast is documented independently. I will look again in January 2027.
Decision Framework: Per-Second Model or Per-Minute Platform
A per-second model generates scenes: B-roll, product shots, and stylized sequences of 5 to 15 seconds. A per-minute avatar platform delivers a presenter reading your script. Teams that need both usually pay for both, so price each job in its own unit: seconds for scenes, minutes for presenters.
For scenes, start from the per-second table above. For presenter video, compare per-minute platforms such as HeyGen's AI video generator, where avatar minutes are billed separately from generated scenes.
FAQ
What does AI video cost per second in 2026?
From about $0.03 to $0.40. HeyGen Video lists $0.03 at 768p, LTX-2.3 Fast $0.04 at 1080p, Veo 3.1 Fast $0.10 at 720p, and Veo 3.1 Standard $0.40, as of October 2026.
What is the cheapest AI video model per second?
HeyGen Video at 768p. It costs $0.03 a second at the standard rate and $0.015 through October 31, 2026. At 1080p, LTX-2.3 Fast at $0.04 is cheaper.
What does Veo 3.1 cost per second?
$0.40 at 720p to 1080p and $0.60 at 4K on Google's Gemini API, with audio included. Veo 3.1 Fast costs $0.10 at 720p and Lite $0.05.
Is Kling or Runway cheaper per second?
Runway, slightly. Runway Gen-4.5 costs $0.12 a second through the API, against about $0.13 for Kling 3.0 through Atlas Cloud. Kling wins if you need 10-second single shots.
Can I still use Sora through an API?
No. OpenAI closed the Sora app on April 26, 2026 and discontinued the Sora API on September 24, 2026.
What is the most common mistake when comparing AI video prices?
Comparing headline rates at different resolutions. A 480p rate and a 1080p rate are different products, so price every model at the resolution you deliver.
Does audio cost extra?
It depends on the host. Google includes audio in Veo 3.1 rates, while fal charges $0.20 a second for Veo 3.1 with audio off and $0.40 with audio on.
The Bottom Line
HeyGen Video has the lowest cost per second in this comparison, $0.03 at 768p with native sound, which puts 1,000 seconds a month at $360 a year. LTX-2.3 is the pick at 1080p and above. Veo 3.1 Fast suits Google Cloud teams. Veo 3.1 Standard is for hero shots. Start with a $5 HeyGen API wallet and price your own clips against the table.
Greetings! My name is Ayesha Shaheryar. My words have helped millions over the past two years. As a HeyGen expert and a writer, I am here to introduce tips and tricks to edit your next video in no time.
