Kling 4.0 API: I Checked 7 Providers - Where Does It Work?

👁 116 reads

Kling 4.0 became available to everyone on October 15, 2026, but only on kling.ai, in apps, and through the official MCP and CLI. There’s still no standard API billed by the second of generated video. That evening, I checked seven places that usually get new video models early and sent two test jobs to fal.ai, which already had Kling 4.0 endpoint URLs. Both returned the same ready-made clip, regardless of the prompt. Here are the results, screenshots, and steps to check for yourself.

Seven providers, seven answers

fal.ai has the endpoints, but no generation

fal.ai published public schemas for seven Kling 4.0 endpoints: text-to-video, image-to-video, references, keyframes, video editing, and two for Flash. The parameters match the release notes:

• duration of 3-30 seconds (Flash: 3-20 seconds); 720p or 1080p resolution (Flash: 720p only);
• auto, 16:9, 9:16, 1:1, and 21:9 aspect ratios;
• multi_prompt - a separate prompt for each shot, with total clip length equal to the sum of the shots;
• keyframes in sequence, up to the duration in seconds plus one, but no more than 10;
• up to 3 voices (voice_ids), referenced in the prompt as @Voice1;
• Flash does not accept a final frame.

I sent two short jobs: a red cube rotating on a white table, 3 seconds, no sound - one to Kling 4.0 Flash and one to full Kling 4.0. Here’s what happened:

The file was byte-for-byte identical, and real video generation takes minutes, not hundredths of a second. This is a placeholder the provider has put in place ahead of the model’s launch. fal’s pricing page also showed placeholder rates (0 USD per second for Flash), so it doesn’t tell us anything either. Bottom line: fal is preparing for the Kling 4.0 API launch, but as of October 15, the model isn’t generating anything there.

MCP and CLI: the only route for automation

Kling’s release notes state: “MCP and CLI now support Kling 4.0, including Omni Reference and video generation with keyframes.” Here’s what I checked without logging in:

• The MCP server at https://kling.ai/mcp returns “401” without a token. Its OAuth metadata points to the kling.ai/auth authorization server and lists scopes for creating generations, reading generations, and checking the credit balance.
• The npm package @klingai/cli-global, version 0.2.1 (dated September 28), is a “thin MCP client”: its commands map one-to-one to the server’s tools. Its instructions list the kling-video-v4_0 and kling-video-v4_0_flash models, keyframe_1..10 frames, and Omni Reference billing per second alongside the length of the input video.

Generations through MCP and CLI draw from your Kling plan credits, not a separate API account. For one person or a small business, this is currently the simplest way to have Claude or ChatGPT request videos in Kling 4.0. I explain the setup step by step in Kling 4.0 MCP and CLI.

How much might the Kling 4.0 API cost?

There’s no Kling 4.0 API pricing yet. For reference, the official Kling 3.0 API pricing on October 15 (1 unit = 0.14 USD) was:

At those rates, a 30-second Kling 3.0 clip in 1080p with audio would cost 5.04 USD. For comparison, I checked Kling 4.0 in the kling.ai creator: 15 credits per second at 1080p, or 0.11 to 0.16 USD depending on the plan (see Kling 4.0 pricing). The 4.0 API rate is unknown - I won’t guess. Kling API plans start at 700 USD for 5,000 units over 180 days.

Until the API can generate video, I don’t have any Kling 4.0 clips of my own. Here’s how the model is presented by its maker:

How to check for yourself

1. OpenRouter’s video model list: GET https://openrouter.ai/api/v1/videos/models - look for an ID containing “kling” and “4”.
2. fal endpoint schema: https://fal.ai/api/openapi/queue/openapi.json?endpoint_id=fal-ai/kling-video/v4/text-to-video (no key required).
3. fal pricing: GET https://api.fal.ai/v1/models/pricing?endpoint_id=... (requires a key). If the unit is “compute seconds” or the price is 0, the model isn’t genuinely available yet.
4. Kling developer page: kling.ai/dev - “Coming Soon” should disappear from there.

I have a script ready that runs the first three checks and submits a test only when the provider lists a real per-second price. Once Kling 4.0 goes live in the API, I’ll run it on the scenes from my Video Arena and add the results to the Kling 4.0 hub.

// read moreKling 4.0 Is Now Available: Features, Pricing and APINewsKling 4.0: Features, Pricing, API and AvailabilityNewsKling Polish Speech Test: Non-English Dialogue QualityBenchmarks

The link to Higgsfield is a partner link: if you buy through it, QUANTSLANT earns a commission and you pay the same price. It has no influence on the content, including where the coverage is critical.