MiniMax H3 vs Veo 3.1
Comparer des modèles
Veo 3.1 is more worth it — 55% vs 46%
How we score worth-it? →Open-weight multimodal video generation with native stereo audio — text/image-to-video, 4–15s clips.
- Running this on my 4070ti super (16GB VRAM), it takes 10 minutes for a 10-second 480p video, but the results are spectacular.— hn ↗
- In the AMA MiniMax said H3 could support sparse attention — that would be a huge speedup. H3 is very cool.— hn ↗
- You landed like Prometheus giving us the power the gods were gatekeeping — I couldn't be more grateful for this immense gift.— huggingface ↗
- Minimax has the best prompt adherence from all open source models I've used — I never got bad prompt adherence.— huggingface ↗
- Being able to run at this level already makes me very happy and satisfied.— blog ↗
- We still do not know the correct objective for preserving its original CFG-distilled behavior — training can drift prompt adherence and temporal dynamics.— huggingface ↗
- When will the face collapse everyone complains about be fixed? If faces keep collapsing, what productivity value does it have?— huggingface ↗
- I can write stuff and the prompt is completely ignored. 99% of videos come out as a TV ad — I think that was the bulk of training data.— huggingface ↗
- Found a small flaw: a blinking problem — characters almost never blink.— huggingface ↗
- 4-bit quantization works, but it's slow to the point of tears. On an RTX PRO 5000, a 540p 15-second I2V takes 700 seconds.— v2ex ↗
“Minimax has the best prompt adherence from all open source models I've used.”
“Precise, high quality, few rerolls, high yield rate.”
“Face collapse that everyone complains about — if faces keep collapsing, what productivity value does it have?”
Slow and underwhelming
- Near the end of 2025, Google was on top — best LLM (Gemini 3 Pro), best video model (Veo 3.1), and best image model (Nano Banana).— hn
- Veo 3.1 still has the edge on raw visual realism for single shots.— hn
- How are they getting such fantastic continuity across clips?— hn
- Even my prompt with Veo3 or Seedance 2.0 generated video can do better than Apple.— hn
- Veo 3.1 produces the highest quality AI video output we have seen. Full stop.— blog ↗
- Also Veo 3.1 is old AF, it can't even do audio references!— hn
- Veo 3 is one of the weaker video models, surprisingly!— hn
- Stuff like veo 3 can't always hold a coherent storyline for 8s.— hn
- The 8-second clip limit is the main constraint. For anything longer, you are stitching clips in post.— blog ↗
- The Standard tier at $0.75/second is hard to justify unless the client is paying for it explicitly.— blog ↗
“The quality when it finally renders is genuinely cinematic.”
“But it's painfully slow. I've waited 10+ minutes for a single clip.”
“Underwhelming for the hype — Google keeps overselling these.”
FAQ
Is MiniMax H3 or Veo 3.1 more worth it?
Veo 3.1 is more worth it, with a 55% worth-it score vs 46% for MiniMax H3.
How much does MiniMax H3 cost vs Veo 3.1?
MiniMax H3: $0.08/s (768P) · $0.13/s (2K) · Veo 3.1: $0.40/sec (audio).
What do users say about MiniMax H3 vs Veo 3.1?
MiniMax H3: Minimax has the best prompt adherence from all open source models I've used.. Veo 3.1: The quality when it finally renders is genuinely cinematic..