whatmodel.sh / video / MiniMax H3

MiniMax H3

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

MiniMax H3 is an open-weights video model from MiniMax, released July 2026. Its published size is 33B parameters. Licence: minimax-h3-community.

78.2±1.4#6 of 93 video modelsfrom 10 benchmark results across 8 sources

$ how it compares

The video models ranked either side of MiniMax H3, out of 139 with a score, then more from MiniMax.

#modelscorewhereRAMreleased
4Wan 3.0 Alibaba79±2api $0.10/sec2026-08
5Dreamina Seedance 2.0 ByteDance Seed79±1api $0.30/sec2026-03
6MiniMax H3 MiniMax78±1api $0.13/sec +local~40GB2026-07
7Dreamina Seedance 2.5 ByteDance77±1api $0.30/sec2026-07
8FLUX 3 Video Black Forest Labs77±2api $0.17/sec2026-08
more from MiniMax
10MiniMax H3 Max MiniMax76±4api $0.080/sec2026-09
48Hailuo 2.3 MiniMax55±2api $0.047/sec2025-10
52Hailuo 02 MiniMax54±2api $0.047/sec2025-06

see all video models · find the ones that fit your machine

$ what it's good at

text to video▆▄▃▄▆▅█▃▂▂▃▄ #7 of 68 * · 37% of its score · 2 tests

Trails FLUX 3 Video, Dreamina Seedance 2.5 here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
LMArena video1460.3 ✎▂▂▃▆█▆·▆▅▁▃▇ beats 35/41avg 1147▇▇▇▇▇▇▇▇▇▇▇▇▇ 34%typical
Sci-VBench testmini (expert human rating)51.6% ✎▅·▃█▃·▅▃▅▃▃▃ beats 8/14hard · avg 44%▇ 3.0%blunt · at its level
image to video▁▂·▂▅█▅▄▅▆▃▄ #2 of 50 · 40% of its score · 4 tests

Beats Gemini Omni Flash, Wan 3.0, Dreamina Seedance 2.0 here, though they rank higher overall.

testresultvs the others in its rundifficultycountswhy
Arena image-to-video1495.4 ✎▁▁▁▁▄▇▃▆▃▂▄█ beats 40/40avg 1189▇▇▇▇▇▇▇▇▇ 22%typical
Rapidata physics realism (I2V, with reference)1339.0 ✎▂··▂▂▆█▆▂▂▆▄ beats 18/19avg 741▇▇▇▇▇ 13%sharp
VGI-Bench visual reasoning (I2V)44.4% ✎▄▄···▄▄█·█·▄ beats 7/8hard · avg 20%▇▇ 4.6%at its level
Rapidata camera movement (I2V)954.0 ✎▄···▄▄▄█··▄█ beats 3/8avg 794▇ 0.6%noisy
video editing▄··█··▄··█▄█ #3 of 9 * · 6.8% of its score · 1 test

Beats Gemini Omni Flash, Dreamina Seedance 2.0 here, though they rank higher overall. Trails Dreamina Seedance 2.5 here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
Arena video edit1392.2 ✎█··█··▄··█▄█ beats 7/9avg 1117▇▇▇ 6.8%typical
other tests3 not in a task · 16% of its score
testresultvs the others in its rundifficultycountswhy
Rapidata relative camera movement (I2V, alignment Elo)1152.9 ✎█···▄▄██·█▄█ beats 7/12avg 661▇▇▇ 7.9%sharp · noisy
Rapidata image-to-video realism (overall Elo)1325.0 ✎▂······▃▂▃▃█ beats 12/12avg 871▇▇ 4.2%sharp · noisy
WROP object-permanence exam (human Elo)1723.6 ✎▄····██···▄█ beats 7/7avg 1299▇ 3.6%noisy

$ where to run it

Through an API

api · fal.ai · $0.13/sec · fal.ai ↗ · price sourceUSD per generated second, fal.ai MiniMax H3 text-to-video at 2K. This is not MiniMax direct API or a reference-to-video tariff. Product: MiniMax H3 text-to-video. API model: minimax/h3/text-to-video. Settings: resolution=2K, input=text only, duration seconds=5–15, audio=not specified by endpoint pricing; no audio-product claim. Price basis: published rate. Currency: USD.

Other ways to run it

version

  1. Run it as published · needs ~51.9 GB · from github.com/leejet
    sd-cli -M vid_gen --diffusion-model models/minimax_h3_fl2va-Q4_K_M.gguf --vae models/minimax_h3_video_vae_fp16.safetensors --audio-vae models/minimax_h3_audio_vae_fp32.safetensors --llm models/qwen3vl_32b_minimax_h3-Q4_K_M.gguf -p "A cute American Shorthair silver tabby kitten surfs on a tropical ocean wave, riding a white surfboard. Cinematic tracking shot, realistic water, bright sunlight, smooth motion. Add upbeat tropical surf-rock background music." --cfg-scale 1.0 -v -W 864 -H 480 --diffusion-fa --offload-to-cpu --rng cpu --fps 24 --video-frames 56

Official weights and documentation: huggingface.co/MiniMaxAI ↗

$ how the score was worked out · LMArena leaderboard dataset 63%, Rapidata/world-model-physics… 13%, Rapidata/relative-camera-move… 8% and 5 more

Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works

sourceresultsshare of the scorelicence
LMArena leaderboard dataset3▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇▇ 63%CC BY 4.0
Rapidata/world-model-physics published dataset-card rankings1▇▇▇▇ 13%CC-BY-4.0
Rapidata/relative-camera-movement-human-preference-900k published dataset-card rankings1▇▇ 8%CC-BY-4.0
VGI-Bench: Probing Visual Intelligence in Video Generation Models1▇ 5%CC-BY-4.0
Rapidata/image-to-video-realism published dataset-card rankings1▇ 4%CC-BY-4.0
Training Object Permanence in World Models (WROP exam; arXiv:2609.28654v1)1▇ 4%CC-BY-4.0
Sci-VBench: Evaluating Knowledge- and Reasoning-Intensive Video Generation in Science Domains1▇ 3%CC-BY-4.0
Rapidata/camera-movement published dataset-card rankings1▇ 1%CC-BY-4.0
prior · pull toward the middle▇ 1%
weighted average= 78.2

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.

data updated · sources and licences · json