Haiper 2.x

Haiper 2.x is the cheapest per-second AI video API on the market at $0.033/sec, now run by NetMind.AI after Haiper's consumer app shut down and its founders joined Microsoft.

Haiper 2.x

Overview

Haiper 2.x is the version bucket the vendor's own API docs use to group the Haiper Video 2.0 and 2.5 checkpoints, a text-to-video and image-to-video model out of London startup Haiper. It launched October 21, 2024 as "Haiper 2.0," a transformer-diffusion hybrid pitched on faster, more realistic generations than the company's earlier 1.0 and 1.5 models. Haiper 2.5 followed on December 18, 2024, adding keyframe conditioning and API integrations with platforms like VEED.

TL;DR

  • Cheapest per-second API in this category at $0.033/sec (540p) and $0.05/sec (720p) - see our video generation pricing comparison
  • Haiper's consumer app shut down in February 2025; the founders joined Microsoft AI and the models were sold to NetMind.AI, which now runs the API
  • No public Artificial Analysis Elo score - Haiper never got tracked on the arena that ranks Kling 3.0 and Veo 3.1

What makes Haiper 2.x worth a model card isn't raw quality - it trails the field there - it's the company's arc. Haiper emerged from stealth in March 2024 with $13.8 million in seed funding (Octopus Ventures, 5Y Capital) on top of a $5.4 million pre-seed round, founded by ex-DeepMind researchers Yishu Miao and Ziyu Wang. By February 2025, less than a year after the 2.0 launch, the consumer web app was gone with no notice to its roughly 4.5 million registered users. The founders joined Microsoft AI in March 2025, and by mid-2025 the video models themselves were sold to NetMind.AI, a decentralized GPU compute provider that now serves "Haiper Text2Video 2.0" through its NetMind Power inference platform. The docs.haiper.ai API and pricing docs are still live and still branded Haiper - that's the version this card covers.

Haiper 2.0 launch graphic showing text-to-video and image-to-video sample outputs alongside the Haiper logo Haiper 2.0's October 2024 launch materials, showing sample outputs from the "Animate your image" and "Create video with text" flows. Source: vp-land.com

Key Specifications

SpecificationDetails
ProviderHaiper (technology operated by NetMind.AI since mid-2025)
Model FamilyHaiper Video (2.0, 2.5 bucketed as "2.x" in official pricing docs)
ParametersNot disclosed
ArchitectureProprietary transformer + diffusion hybrid
Clip LengthUp to 6 seconds (4s or 6s presets via API)
Resolutions (video, priced)540p, 720p only - no 1080p or 4K video tier in current pricing docs
Native AudioNo
API Price$0.033/sec (540p), $0.05/sec (720p)
Release DateOctober 21, 2024 (2.0), December 18, 2024 (2.5)
LicenseProprietary, API-only

Benchmark Performance

Haiper has never appeared on the Artificial Analysis Video Arena, the blind-preference Elo leaderboard that ranks Kling 3.0, Veo 3.1, and Runway Gen-4.5 against each other - so there's no independent Elo score to report, and I'm not going to invent one. What's verifiable instead is API-side positioning: price per second, shipped resolution, and clip length against the competitors it's actually priced to undercut.

ModelProvider720p API Price/secMax Video ResolutionMax Clip LengthNative AudioArtificial Analysis Elo
Haiper 2.xHaiper / NetMind$0.05720p (4K "coming soon")6sNoNot ranked
Runway Gen-4 TurboRunway$0.054KNot disclosedNoNot disclosed
LTX-2 FastLightricks$0.04 (1080p)4KNot disclosedNo1,127
Kling 3.0 1080p ProKuaishou$0.075Native 4K/60fps15sYes (multilingual)1,251 (no audio)
Veo 3.1 FastGoogle DeepMind$0.104K8sYes1,205

The headline number is that Haiper's 720p rate is identical to Runway Gen-4 Turbo's $0.05/sec - but Runway ships true 4K output and Haiper doesn't. Haiper only undercuts on the 540p tier, where $0.033/sec is roughly a third of Runway's flat rate. Against Kling 3.0 and Veo 3.1, Haiper is cheaper across the board, but neither of those models is competing on price - they're competing on the resolution, duration, and audio Haiper lacks. See the video generation benchmarks leaderboard for the full Elo standings this table draws from.

Key Capabilities

Text-to-video and image-to-video

Haiper 2.x's core endpoints are straightforward: submit a text prompt or a source image and get back a 4- or 6-second clip. Image-to-video pricing matches text-to-video at the same resolution tier, which is standard across the category. There's no native video-to-video or reference-based generation in the current API - features like Kling's Omni reference-to-video or Veo's Ingredients to Video simply don't have a Haiper equivalent.

Video extension and repainting (legacy 1.5 endpoint)

The pricing docs list a video extension endpoint and a 1080p upscale endpoint, but both are scoped to the older 1.5 model, not 2.x, at $0.05/sec each. If you're building on the 2.x endpoints specifically, extension and upscaling aren't available to you without mixing model versions - a gap worth checking before committing a pipeline to it.

The 4K promise that still hasn't shipped

Haiper's October 2024 launch marketing said 4K resolution was "coming soon." As of the current docs.haiper.ai/api-reference/api-pricing page - nearly two years later - video generation pricing only lists 540p and 720p tiers; 4K video is still flagged as a future feature. (4K is live for Haiper's separate image-generation endpoint, at $0.12/image, which makes the video gap more conspicuous rather than less.) That's the kind of marketing-versus-shipped-product gap worth flagging before you plan a resolution-dependent workflow around it.

NetMind.ai branded content showing the halftone dot pattern used across Haiper's marketing materials Haiper 2.0's launch visual identity, since carried into the NetMind Power-hosted version of the API. Source: prnewswire.com

Pricing and Availability

The current commercial surface for Haiper 2.x is API-only, accessed at api.haiper.ai with docs at docs.haiper.ai. There's no consumer web app anymore - haiper.ai's product pages that once listed Free/Explorer/Pro subscription tiers (referenced in older third-party reviews) 404 or redirect today, consistent with the February 2025 consumer shutdown. Treat any "$8/month Explorer" or "$24/month Pro" figures you see on review-aggregator sites as historical, not current.

TierPriceNotes
Video, 540p$0.033/secText-to-video and image-to-video
Video, 720p$0.05/secText-to-video and image-to-video
Video extension (1.5 only)$0.05/secNot available on 2.x endpoints per current docs
Upscale to 1080p (1.5 only)$0.05/secNot available on 2.x endpoints per current docs
Image, 720p$0.04/imageHaiper Image 2.x
Image, 1080p$0.08/imageHaiper Image 2.x
Image, 4K$0.12/imageHaiper Image 2.x

A 6-second clip at 540p costs $0.20; at 720p it's $0.30. That's cheaper than a comparable clip on Runway Gen-4.5 ($1.50 at $0.25/sec) or Veo 3.1 Fast, and it's the reason Haiper still shows up as the "cheapest API" line item in our own AI video generation pricing comparison. Access today runs through the standard API key flow documented at docs.haiper.ai; there's no published subscription plan, credit system, or free tier in the current developer docs, a change from the consumer-app era. For teams evaluating tools rather than raw API integration, our best AI video generation tools roundup covers the platforms - Runway, Kling, Pika, HeyGen - that replaced Haiper as the default consumer pick after the shutdown.

Strengths and Weaknesses

Strengths

  • Cheapest per-second video generation API in the category at $0.033/sec (540p)
  • Image-to-video priced identically to text-to-video - no premium for conditioning on a source image
  • Simple, documented REST API still actively serving requests despite the parent company's consumer shutdown
  • Image generation endpoints (separate from video) do ship real 4K output at $0.12/image

Weaknesses

  • No public benchmark presence - not tracked on the Artificial Analysis Video Arena that ranks every major competitor
  • Video output caps at 720p in the current pricing docs; the 4K video tier promised at the October 2024 launch still isn't live
  • No native audio, and the video extension/upscale endpoints only apply to the older 1.5 model, not 2.x
  • 6-second maximum clip length is short next to Kling 3.0's 15 seconds or Veo 3.1's 8 seconds
  • Consumer product is dead; anyone relying on the old web app, subscription tiers, or saved cloud projects lost access completely in February 2025
  • Company continuity risk - the founding team left for Microsoft and the technology changed hands to NetMind.AI within about a year of the 2.0 launch

FAQ

Is Haiper AI still available?

The consumer web app shut down in February 2025. The underlying Haiper Video 2.x model is still accessible as a developer API at docs.haiper.ai, now operated through NetMind.AI's infrastructure after the June 2025 acquisition.

What happened to Haiper's founders?

Cofounders Yishu Miao and Ziyu Wang, both former Google DeepMind researchers, joined Microsoft AI in March 2025 along with machine learning researcher Edward Hayes, reporting to Nando de Freitas.

Why is Haiper 2.x so much cheaper than Runway or Veo?

Haiper trades resolution, duration, and audio for price. At $0.033/sec (540p) it's roughly a third of Runway Gen-4 Turbo's flat rate, but it caps at 720p and 6-second clips with no native audio, versus Runway's 4K or Veo 3.1's synchronized audio track.

Does Haiper 2.x support 4K video?

No. Despite marketing 4K as "coming soon" at the October 2024 launch, the current API pricing docs only list 540p and 720p video tiers. Haiper's separate image-generation endpoint does support 4K at $0.12/image.

Is there a free tier for Haiper's API?

The current developer documentation doesn't list a free tier or credit system - it's straight per-second API pricing. Free-tier and subscription details found on third-party review sites describe the discontinued consumer web app, not the current API.

Sources

✓ Last verified July 20, 2026

James Kowalski
About the author AI Benchmarks & Tools Analyst

James is a software engineer turned tech writer who spent six years building backend systems at a fintech startup in Chicago before pivoting to full-time analysis of AI tools and infrastructure.