MiniMax H3 and Music 3.0: two new multimodal models
MiniMax has shipped two new multimodal models back to back: H3 for video (launched 7/31, open-sourced 8/3) and Music 3.0 for music (launched 8/13), both already billing pay-as-you-go on the API. H3 is a general multimodal video model that outputs up to 2K, 24fps, 15-second clips with native stereo audio (source); Music 3.0 turns a brief idea plus optional lyrics into a complete song up to 5 minutes, covering composition, arrangement, vocals and production (source).
For you this means two new modalities to call: short-form video, ad storyboards and trailer previews go to H3; finished songs, covers and instrumentals go to Music 3.0. Just remember the billing units differ — one charges per second, the other per song — so keep them separate when doing a pricing comparison.
Quick pricing comparison
| Model | Billing unit | Core price | Notes |
|---|---|---|---|
| MiniMax H3 (2K) | second | ¥0.80/sec | 2K video generation |
| MiniMax H3 (768P) | second | ¥0.50/sec | 768P video generation |
| MiniMax Music 3.0 | song | ¥1/song | Full song, free tier limited |
Prices come from MiniMax's pay-as-you-go docs (source).
MiniMax H3 video pricing comparison
H3 pricing splits by resolution plus per-feature add-ons, not a single flat rate. The pay-as-you-go page itemizes 2K, 768P, Regeneration and Context-IR separately (source):
| H3 item | Price | Notes |
|---|---|---|
| 2K generation | ¥0.80/sec | up to 2K output |
| 768P generation | ¥0.50/sec | default 768P output |
| Regeneration (768P→2K) | ¥0.30/sec | upscale 768P to 2K |
| Context-IR | ¥5.80 in / ¥23.00 out per 1M tokens | multimodal instruction understanding |
| Input images | first 5 free, ¥0.20/image after | first/last-frame or reference inputs |
The pricing comparison conclusion is direct: 2K costs 60% more than 768P for higher detail, good for deliverables; 768P is the value tier for drafts and quick tests. And the official "full 2K workflow" actually stacks 768P generation plus Regenerate-to-2K, so the real cost is higher than the ¥0.8/sec 2K sticker.
MiniMax Music 3.0 pricing comparison
Music 3.0 is the simplest to price: per song, ¥1 for the standard model, with the free variant currently at no charge. The docs split it into standard, free, and lyric generation (source):
| Music 3.0 item | Price | Notes |
|---|---|---|
| Music 3.0 standard | ¥1/song | full song generation |
| Music 3.0 free | limited free | currently no charge |
| Lyric generation | ¥0.05/song | standalone lyrics |
Compared to H3, Music 3.0 is friendlier for creators: a song is ¥1, so even several revisions stay cheap. The free tier has no published end date, so budget for production on the ¥1/song standard instead of treating the promo as permanent.
Pricing comparison against MiniMax's own models — worth it?
Dropping H3 and Music 3.0 into MiniMax's own lineup, they fill the video and song gaps rather than replacing the M3/M2.7 text models. The API now spans text, speech, video, image and music (source):
| Model | Billing | Core price | Best for |
|---|---|---|---|
| MiniMax-M3 | per 1M tokens | ¥2.1/¥8.4 (≤512K, permanent 50% off) | complex agents, long-context code |
| MiniMax-M2.7 | per 1M tokens | ¥2.1/¥8.4 | production text/agents |
| Speech-2.8 | per 10K chars | ¥2-3.5/10K chars | TTS |
| MiniMax-H3 | per second | ¥0.50-0.80/sec | video |
| Music-3.0 | per song | ¥1/song | songs/covers/instrumentals |
| image-01 | per image | ¥0.025/image | batch image generation |
On pricing comparison, whether it's worth it depends entirely on your main workload: agents and code use M3/M2.7, voiceover uses Speech-2.8, video uses H3, music uses Music 3.0. The units all differ, so don't compare M3's ¥2.1 directly against H3's ¥0.8 — different units, no direct comparison. For cross-vendor video and music pricing, also check the Volcengine, Alibaba Qwen and iFlytek comparison pages; H3 positions as general multimodal video and Music 3.0 as production-grade songs, so which is cheaper depends on your resolution, clip length and output volume.
API key, integration, quota and rate limits
H3 and Music 3.0 both bill via a standard API key, separate from the MiniMax Token Plan credit pool. Key points:
- API keys are created in the open-platform console and deduct from account balance; Token Plan subscription keys use the credit pool, and the two don't mix (source).
- Integration uses OpenAI/Anthropic-compatible APIs with Cache and Agent support; H3 video and Music 3.0 run under the same API, so no separate key is needed.
- Usage is driven by balance and per-model pricing — H3 accrues per second, Music 3.0 per song; fixed rate limits aren't published, so check the console.
- High-volume video has prepaid packs from ¥7,000 (credit pool), and speech has HD packs from ¥630 and Turbo from ¥360 — a third system separate from pay-as-you-go balance and Token Plan, so don't top them up interchangeably.
- For a fixed monthly fee and high-frequency agents, use the MiniMax Token Plan; use a standard API key for pay-as-you-go video and music.
Things to watch
-
H3's 2K full workflow stacks cost. The official "768P generate + Regenerate to 2K" path is effectively ¥0.5/sec + ¥0.3/sec, not just the ¥0.8/sec 2K sticker.
-
Music 3.0 free tier is not permanently free. Only "limited free" is stated with no billing timeline, so budget production on ¥1/song.
-
Open-source H3 doesn't remove the API entirely. H3-Base (768P) runs locally, but the 2K workflow's Context-IR and Regenerate-2K remain official API, so high-quality 2K still needs top-up.
-
Don't mix API and Token Plan keys. A standard API key bills balance while a subscription key bills credits — different billing and quota — so confirm which one you're wiring up before integration.