Kling 4.0 vs Hailuo 2.3: A Six-Month Gap in Two Directions

There is a version of this comparison that writes itself: Kling 4.0 launched on 15 October 2026, Hailuo 2.3 arrived in April, so the newer model wins. That version is wrong in a way that costs money, because the two models were built for different jobs and the six-month gap has widened the difference rather than closed it. Hailuo 2.3 is a stylist with native 1080p and no audio. Kling 4.0 is a controller with 30-second takes, stereo sound and ten keyframes. If you generate anime-adjacent motion work and cut music over it in an editor, the older model is the better instrument and always was. If you need a twenty-second take with dialogue that syncs, Hailuo 2.3 cannot do the job at any price. The documented AI failures and agent incidents behind releases like this one are collected in the OrcaRouter AI Incident Archive where each entry carries its own sources, severity, confidence grade and disputed flag.

Which is why the useful question is not "which is better" but "which failure can you live with." Below is where each one fails.

What Hailuo 2.3 actually is

M​iniMax's model is a stylist, and the company is upfront about it: cinematic camera motion, human motion and micro-expressions, and stylised or anime content are the strengths, and photorealism is explicitly not the target. That is a real market — a large share of short-form video is stylised — and on that ground it competes with models that cost several times more.

The spec sheet is where the age shows. Resolution is native 1080p, and unlike several rivals that upscale, M​iniMax does not attach an upscale caveat to it. Clip length is the constraint: 6 seconds at 1080p, 10 seconds at 768p, and no end-frame control, which makes multi-shot continuity a manual stitching job. There is no native audio — an earlier Hailuo generation had it and 2.3 dropped it — so every clip needs sound added downstream.

• Resolution — Hailuo 2.3 native 1080p vs Kling 4.0 720p, 1080p and 4K, with 10-bit HDR at the top two marked coming soon.

• Clip length — Hailuo 2.3 6 seconds at 1080p, 10 seconds at 768p vs Kling 4.0 3 to 30 seconds in a single pass, with 2-minute extension listed as coming soon.

• Audio — Hailuo 2.3 none vs Kling 4.0 two-channel stereo, lip sync and dialogue in nine languages.

• Frame control — Hailuo 2.3 start frame only, no end-frame pinning vs Kling 4.0 up to 10 keyframes plus first-and-last-frame.

• Reference control — Hailuo 2.3 not positioned as a reference-first model vs Kling 4.0 up to 15 combined reference assets.

• Commercial shape — Hailuo 2.3 prepaid points packages from $1,000 to $6,000 vs Kling 4.0 subscription credits in the app, no published API rate.

• Independent score — neither model has a current head-to-head; Hailuo 2.3's last board reading was around #28 at roughly Elo 1,180 on an earlier edition of the Artificial Analysis video board, and Kling 4.0 is absent.

The price comparison nobody can finish

Hailuo 2.3's official channel is prepaid points: packages running from $1,000 to $6,000, burning at a fixed rate — one point for a six-second 768p standard clip, two points for a ten-second 768p or six-second 1080p clip, with the faster Hailuo-2.3-Fast variant dropping those to about 0.7 and 1.1 to 1.3. Resellers quote a flatter $0.067 to $0.08 per second, which is where the "Hailuo is dirt cheap" reputation comes from.

Kling 4.0's API has no published rate. The model sits behind an early-access banner in K​uaishou's developer documentation, the pricing tables list Kling 3.0 and its tiers but not the new model, and the consumer product runs on subscription credits. Reseller listings exist for Kling 4.0 endpoints, but a reseller rate is not a vendor list price and does not tell you what the model will cost when the endpoint opens.

So the honest statement about cost is this: the cheaper model is Hailuo 2.3, because it is the only one of the two with a price. That is not a compliment to its economics — a solo creator comparing a $1,000 minimum commitment against an unknown is not comparing rates at all. It is a statement about which side of the comparison has been de-risked.

The audio question settles more than the spec sheet does

If the deliverable needs sound generated with the picture — dialogue, ambience, a voice in a specific accent — Hailuo 2.3 is out before the comparison starts. Not "weaker": out. There is no audio channel to evaluate. The workflow is generate silent, then score and d​ub in an editor, and if the shot has a speaking character the lip sync must be fixed by hand or by a separate tool.

Kling 4.0's whole launch is built around the opposite assumption. Two-channel stereo with tighter lip sync, dialogue across nine languages including named dialects, and on-screen text generation that is claimed to stay legible as the camera moves. For a talking-head clip, a short drama scene or a product film with a voice-over baked in, that is the capability that decides the purchase, and it is the one thing Hailuo 2.3's documentation does not dispute.

The counter-argument is that generated audio is still the weakest part of every video model, and K​uaishou's own material puts no benchmark behind the lip-sync claim. That is true and it is why the audio row should be read as "can do the job at all" rather than "does it well." For a first pass, being able to generate a synced take and fix it is different from having to build the sound from zero.

Where the stylist still wins

Two places, and they are worth naming precisely because the newer model's advantages are so much easier to list.

The first is native resolution at the price. Hailuo 2.3 gives 1080p without an upscale caveat and without a "coming soon" attached to the colour pipeline. Kling 4.0's 4K is a shipped output setting but its 10-bit HDR is not, so for a colour-managed workflow at 1080p the older model is the cleaner source — no surprises, no roadmap dependency.

The second is the look. A model trained and tuned toward stylised motion produces anime-adjacent and graphic work that a generalist model has to be coaxed into, and K​uaishou's release note lists claymation, glitch, felt and 3D cyber styles as supported looks without claiming any of them as a speciality. If your format is stylised short-form, the model that was built for it will beat the model that lists it among twenty options, and no keyframe count changes that.

There is a third, softer point. Hailuo 2.3's weaknesses are documented by its own vendor — no audio, no end frame, not photoreal — which makes it easy to plan around. Kling 4.0's release note names no weakness at all; its only hedges are the coming-soon labels. A model with a published list of what it is bad at is easier to put in a production plan than one with a published list of what it will be good at.

Why Hailuo 2.3 is the wrong M​iniMax model to compare against

One thing this matchup hides: Hailuo 2.3 is not M​iniMax's current generation. MiniMax H3 arrived on 31 July 2026 as an omni-modal model that reads text, image, video and audio references as one context, generates 4 to 15 second clips at 768p or 2K with native stereo audio, and is billed per second of output. Comparing Kling 4.0 to Hailuo 2.3 is comparing October 2026 to April 2026 and skipping the model in between that actually has audio, references and a modern duration ceiling.

That matters for the decision because H3 is the model a Kling 4.0 buyer would realistically weigh — and it is the one we route. On OrcaRouter, minimax/minimax-h3 is listed at $0.08 per second of generated output, on the same OpenAI-compatible endpoint as 200+ other models, at provider list price with 0% markup. Hailuo 2.3 is not in our catalogue, and neither is Kling 4.0; the K​ling family we carry runs from the 2.5 generation through Kling 3.0 and Kling 3.0 Omni at $0.084 per request and Kling 3.0 Turbo at $0.112. So the practical routing question in October 2026 is Kling 4.0 against H3, with Hailuo 2.3 relevant only if you already hold M​iniMax points or specifically need its look.

The one-sentence test

If you can describe your shot without mentioning sound, and the look is stylised, Hailuo 2.3 is still the right tool and its age is irrelevant — 1080p, native, no upscale, at a rate you can negotiate down at volume. If sound is part of the shot, or the shot is longer than ten seconds, or you need to pin ten moments on a timeline and edit one of them afterwards, Kling 4.0 is the only one of the two that can attempt it — in the app today, with the API to follow.

And if you are choosing for a pipeline rather than a project, compare Kling 4.0 against MiniMax H3 rather than Hailuo 2.3, because that is the pairing where both sides have audio, both sides have references, and one of them has a published per-second price. The three datable things to watch are Kling 4.0's API leaving early access with a rate, Kling 4.0's first independent Elo, and whether M​iniMax keeps Hailuo 2.3 on sale beside H3 or retires it — because a model with no successor pressure is a model whose price stays put.