Kling 4.0 vs Flux 3 Video: One Has a Rate Card, the Other Has a Ceiling

Both Kling 4.0 and Flux 3 Video accept text, images and video, both promise 4K output, and both generate audio in the same pass as the picture. The difference is that you can price one of them. Flux 3 Video, from Black Forest Labs, went to general availability on 4 August 2026 with a published per-second rate card and a scored row on the Artificial Analysis text-to-video board. Kling 4.0 has been in the consumer app since 15 October 2026, generates up to 30 seconds in one pass, and has neither a vendor per-second rate nor an independent Elo. So the comparison is not simply feature against feature; it is a product you can budget against today against a product whose ceiling is higher on paper.

That makes the decision unusually clean if you separate two questions. Which model can do the shot? And which model can you actually commit to this quarter? The answers are not the same, and most head-to-head pages collapse them into one.

Two boards, no shared column

Artificial Analysis lists Flux 3 Video on the text-to-video board, sixth at Elo 1,128 with a 95% confidence interval of ±9 over 6,179 samples, at a normalized $17.40 per minute, with a release date recorded in July 2026. Black Forest Labs dates the public release to 4 August 2026, after early access from 23 July; the discrepancy is a board-record artifact, not a contradiction about the model.

Kling 4.0 is absent from that board and from the image-to-video and editing boards as of 23 October 2026. The previous generation, Kling 3.0, has rows; the new model does not. A comparison page that puts an Elo next to Kling 4.0 is quoting the old model. The only honest statement is that one side of this matchup has a measured position and the other has a blank.

Flux 3 Video's row is also narrower than its product. The vendor documents text-to-video, image-to-video, video-to-video and video continuation, with keyframe-controlled transitions. Artificial Analysis scores only the text-to-video configuration, so the board number describes one mode of a multimodal model, not the whole surface. That is a limit of the scoreboard, not of the model.

What goes in

Flux 3 Video is a multimodal video model with a broader input list than its board row suggests: text prompts, still images as a first frame, video for video-to-video and continuation, and keyframes for controlled transitions. Duration runs up to 20 seconds in a single generation, with 5 to 20 seconds for text-to-video and image-to-video, and 5 to 15 seconds for continuation. Resolution is sold in pixel bands rather than fixed aspect ratios — HD at roughly 1 megapixel per frame, FHD at 2, QHD/2K at 4 and UHD/4K at 8 — and the vendor offers optional native audio with multilingual dialogue, synchronized speech, sound effects and ambience, included at no extra charge.

Kling 4.0's input story is built around a budget rather than a mode list. Omni Reference takes up to 15 assets in one task, capped at 10 images, 5 videos totalling 30 seconds, and 7 subjects of which up to three may be video-based; each subject can carry a specified voice. On top of that it accepts up to 10 keyframe images, prompts up to 8,000 tokens, and outputs 3 to 30 seconds at 720p, 1080p or 4K, with 21:9 among the aspect ratios. The release note also lists targeted editing across up to five input clips within a 30-second combined window.

The contrast is not “one takes images and the other does not.” It is that Flux 3 Video exposes inputs as distinct modes with distinct durations, while Kling 4.0 exposes them as one task with a shared allowance. A 20-second Flux generation and a 30-second Kling generation are both single passes; the Kling pass can carry more identity and motion references, and the Flux pass has a published price.

Resolution and the HDR caveat

Both vendors advertise 4K, and both deserve the same scrutiny. Black Forest Labs sells resolution as a pixel band and lists HD, FHD, QHD/2K and UHD/4K as rate-card tiers; several third-party summaries report that 2K and 4K are announced or rolling out rather than universally available, so treat the top bands as a roadmap question at your account level rather than a default. Kling 4.0 lists 720p, 1080p and 4K as output settings, but 10-bit HDR at 1080p and 4K is explicitly marked “coming soon” on Kuaishou's release note and feature page. Native 4K without HDR is the shipped claim; HDR is not.

Neither model publishes an independent resolution-normalized quality comparison. The boards normalise price to a minute of 1080p, which is why Flux 3 Video's $17.40 per minute is comparable to other rows but not to Kling 4.0's missing number.

Price: the part that actually decides this quarter

Flux 3 Video's rate card is specific. Text-to-video and image-to-video run $0.06 per second in draft HD, $0.17 in HD/720p, $0.29 in FHD/1080p, $0.40 in QHD/2K and $0.80 in UHD/4K. Video-to-video and continuation are higher: $0.12 draft, roughly $0.41 to $0.43 HD, $0.53 to $0.54 FHD, $0.65 2K and $0.95 4K. Audio is included. A five-second HD clip costs about $0.85 at the standard rate, before any volume or platform arrangement.

Kling 4.0 has no vendor per-second rate as of 23 October 2026. The app runs on credits through Kling AI's subscription plans. The API is behind an early-access banner, and Kuaishou has not published a rate card. Third-party platforms list per-second figures, but those are reseller prices, not the vendor's, and they move independently. A cost comparison between the two models today would be a comparison between a published rate and an estimate.

For a usable baseline, the previous Kling generation is priced. On OrcaRouter, kling/kling-v3 and kling/kling-v3-omni are listed at $0.084 per request, kling/kling-3-turbo at $0.112, and the older 2.x tiers at $0.042 to $0.28. Those are per-request catalogue prices rather than per-second rates, so they do not normalise directly against Flux's card, but they are real numbers you can put in a budget while Kling 4.0's rate is unknown.

Audio and the track-count question

Flux 3 Video's audio is optional, native and included in the per-second price, with multilingual dialogue, synchronized speech, sound effects and ambience. Kling 4.0 generates two-channel stereo, with improved lip sync and dialogue across nine languages plus named English and Mandarin accents. Both are vendor claims, and neither vendor publishes a track count. If you need dialogue, music and effects as separable stems, neither spec sheet tells you whether you will get them. Budget for a second pass in a DAW or a regeneration, not for a clean multitrack export.

Weights, access and integration

Neither model is open weights. Flux 3 Video's board row carries no weights link, and Black Forest Labs serves it through its API and dashboard; Kling 4.0 is a hosted model through Kuaishou's app and open platform. If self-hosting is a requirement, neither is the answer, and Black Forest Labs' open-release reputation does not change the licence on this checkpoint.

Flux 3 Video has been generally available since early August, so its request contract has had a quarter to settle. Kling 4.0's API is in early access, which historically means an application and a wait rather than a public rate card; MCP and CLI support are listed, which helps agents and build scripts but does not make the endpoint generally available.

Neither model is an OrcaRouter route today, and this article is not a claim that we host either. What we do carry is the Kling 3.0 family and more than 200 text and multimodal models on one OpenAI-compatible endpoint, at provider list price with 0% markup. For a team piloting Kling 4.0 while keeping a production path on Flux or Kling 3.0, the useful architecture is a stable model name in front of the call and automatic failover behind it, so an early-access endpoint that rate-limits or disappears does not take the batch with it. That is the same argument for holding two video vendors at all: the second one is insurance, not a hedge on quality.

Who should pick which

• Pick Flux 3 Video if you need a published per-second price, a scored text-to-video row, a 20-second ceiling and native audio included. It is the only one of the two you can approve in a budget meeting this week.

• Pick Kling 4.0 if the shot needs 30 seconds without a cut, more than a handful of identity references, or a separate keyframe track. The ceiling is higher and the price is unknown.

• Pick Flux 3 Video for video-to-video and continuation with a known rate; Kling 4.0's editing is documented on clips it generated, within a 30-second combined window, not as a general footage-editing mode.

• Pick Kling 4.0 for a 21:9 deliverable with 10 keyframes and a 15-asset cast. Check HDR expectations at the door.

• Pick neither if separable audio stems are a hard requirement until one vendor documents the channel architecture.

What would settle this is a Kling 4.0 rate card and a first Artificial Analysis row. Until both exist, Flux 3 Video is the safer purchase and Kling 4.0 is the more capable pilot. The gap between those two sentences is the entire decision.