MiniMax prices H3, the model the community calls Hailuo 3, per second of output: $0.08 at 768P and $0.13 at native 2K on its own platform, with the first five reference images free and $0.04 for each image after that. Replicate passes that list price through; fal is lower at 768p ($0.06); Kie.ai lists half of official, $0.04 and $0.065.
On APIMODELS there are two ways to run it, and they sit at opposite ends of this table. minimax-h3 is the full tier with native 2K and reference video, at $0.097 per second at 768P and $0.145 at 2K, above official and above most resellers. minimax-h3-lite runs the H3 model at 768P for $0.02 per second and at 480P for $0.01, the lowest price for H3 we found anywhere, with one real trade-off described below.
USD per second of output, checked 2026-09-30. Promotions are marked and not treated as list prices.
| Provider | 480P | 768P | 2K | Notes |
|---|---|---|---|---|
| MiniMax official | — | $0.08 | $0.13 | 5 reference images free, then $0.04 each |
| Replicate | — | $0.08 | $0.13 | 4–15 s |
| EvoLink | — | $0.076 | $0.124 | official × 0.95 |
| fal | $0.05 | $0.06 | $0.13 | 4K $0.16 |
| Kie.ai | — | $0.04 | $0.065 | half of official |
| APIMODELS minimax-h3 | — | $0.097 | $0.145 | reference video supported |
| APIMODELS minimax-h3-lite | $0.01 | $0.02 | — | draft tier, 1–15 s |
H3 Lite is cheap because it is a draft tier, and it behaves like one on the hardest shots. On animation, talking heads, slow camera moves, product turns and subtle expressions it holds up; on live-action footage with fighting, running, dancing or fast camera work, faces and limbs deform, and rewriting the prompt does not fix it. It also stops at 768P, takes no reference video, and a clip takes several minutes. Over the last 30 days of external traffic it served 661 calls at a 99.2% success rate, so it is dependable for what it is.
A practical split: iterate on prompts with H3 Lite at 480P for a cent a second, then render the chosen shot on minimax-h3 or at MiniMax directly when it needs 2K or large movement.
cURL
curl -X POST https://api.apimodels.app/v1/video/generations \
-H "Authorization: Bearer $APIMODELS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "minimax-h3-lite",
"prompt": "A cartoon fox waves at the camera, slow push-in, warm light",
"resolution": "480p",
"duration": 6
}'Python
import requests
H = {"Authorization": "Bearer YOUR_APIMODELS_KEY"}
r = requests.post("https://api.apimodels.app/v1/video/generations", headers=H, json={
"model": "minimax-h3", # full tier: 2K and reference video
"prompt": "The dancer from video 1 performs the same move on a rooftop at sunset",
"reference_video_urls": ["https://example.com/move.mp4"],
"resolution": "2k", "duration": 8,
})
print(r.json()) # poll GET /v1/video/generations?task_id=...On MiniMax’s own platform H3 is $0.08 per second at 768P and $0.13 at 2K. fal charges $0.06 at 768p, Kie.ai $0.04 and $0.065. On APIMODELS, minimax-h3 is $0.097 and $0.145, and minimax-h3-lite runs H3 at $0.02 per second at 768P and $0.01 at 480P.
Yes. H3 is MiniMax’s model name; Hailuo is the consumer app, and "Hailuo 3" is what most people search for. OpenRouter lists it as hailuo-3.
For final live-action shots with large or fast movement, for anything above 768P, and for jobs that need a reference video. Use minimax-h3 or MiniMax directly for those.
