AI Video Comparison Correction: Unsupported Durations Withdrawn, Sora API Removed

Correction, October 15, 2026: this is the corrected archive of our July 19 comparison, not a current five-product benchmark. We did not record a reproducible common-task test. The former generation-time, maximum-duration, price and quality-leadership assertions lacked adequate sources and are withdrawn, including the recommendations that relied on them. Use our current video workflow guide for a documented trial plan.

What this archive can establish#

The original article grouped Kling, Sora, Runway, Vidu and Hailuo by intended creative task. That editorial grouping did not establish which produced better footage, generated faster or cost less on the same task. We no longer present the former Vidu speed or Runway long-duration assertions as reasons to buy either product. We also withdraw the uncited monthly price range and claims of a uniquely best-value tool.

Sora: the retired scope is specific#

OpenAI's deprecation table, read October 15, lists the Videos API and named sora-2 aliases and snapshots as removed on September 24, 2026, with no recommended replacement. That removes this API/model scope from a current shortlist. It does not prove a standalone-app shutdown date or a move into ChatGPT; we withdraw those unsupported extensions. The Sora directory record states the same boundary.

How to use the remaining product names#

Treat Kling, Runway, Vidu and Hailuo as names to investigate, not four verified equivalents. Start with your deliverable, check the exact model and entry point, obtain a current quote and the applicable commercial-use terms, then compare the same authorised source assets. We have not re-verified every remaining product's current availability or licence here. The current guide separates readable evidence from conditional candidates.

For e-commerce, inspect whether packaging text and product identity survive a shot. For an ad, inspect continuity across approved shots. For a talking-head clip, inspect voice/likeness permission and synchronisation. These are proposed acceptance criteria, not results we measured or product winners.

Method and correction record#

This is documentary and editorial research. No shared model, plan, prompt set, runtime or quality scores were recorded, so the former language of supremacy and precise speed comparisons was not justified. Commercial-use permission cannot be inferred from a free tier or a paid subscription. See our editorial standards.

October 14's note had labelled numbers unknown while leaving the recommendations that depended on them. This October 15 correction removes both the unsupported assertions and their downstream advice; it also limits Sora statements to the evidenced API/model scope. The original event date and URL are retained.