Minimax H3 vs Veo 3.1: Arena Leader vs Ecosystem

Last verified 2026-10-07 · Comparison · How we compare · Evidence
Higher on both LMArena boards

MiniMax H3

MiniMax

MiniMax's general-purpose multimodal model: clips of up to 15 s[1] at 2K resolution with native stereo sound[2], billed per second on the MiniMax platform API from $0.08/s[3] at the 768P tier[3]. It ranks first[4] on LMArena's image-to-video leaderboard with a score of 1495[4], and open-weight H3-Base checkpoints are published on Hugging Face[5].

4K, extensions and a Lite tier

Veo 3.1

Google

Google's current video model: 8 s[6] clips that extend to 148 s[6] at 720p, 4K output on 8-second clips[6] and native audio[7], billed per second in the Gemini API[8].

Quick verdict

Pick MiniMax H3 when ranked quality is the brief: it sits above Veo 3.1 on both LMArena boards (1460[9] against 1364[10] on text-to-video, 1495[4] against 1398[11] on image-to-video), generates 15 s[1] in one pass against 8 s[6], and accepts 12[1] reference files against 3[6] images. Pick Veo 3.1 for the price floor and the ceilings H3 lacks: the Lite tier starts at $0.05/s[8] against H3's $0.08/s[3], 4K output exists on 8-second clips[6], and extensions build a video out to 148 s[6] at 720p.

Category winners are computed from the sourced rows in the tables below; “Depends” means there was not enough comparable public data to call it.

Summary

MiniMax launched H3 on 2026-07-31[2] with native stereo audio and multi-shot modeling[2]; Google's Veo 3.1 reached the Gemini API on 2025-10-15[12] and remains a paid preview there[12]. Both generate sound with the picture and both run at 24 frames per second[5][6].

The spec sheet favours H3 on working room: 15 s[1] per generation against 8 s[6], and up to 12[1] mixed reference files (9 images, 3 videos and 3 audio clips[1]) against 3[6] reference images. Veo claws back on ceilings: 1080p and 4K renders on 8-second clips[6] where MiniMax states a 2K maximum[2], plus extensions of 7 seconds at a time out to 148 s[6] at 720p; we found no extension feature in H3's documentation.

Pricing splits by tier rather than by model. H3 charges one flat rate per resolution: $0.08/s[3] at 768P and $0.13/s[3] at 2K[3]. Veo 3.1 spans $0.05/s[8] (Lite, 720p) to $0.60/s[8] (Standard, 4K), so its cheapest tier undercuts H3 while its flagship tier costs five times H3's base rate ($0.40/s[8] at 720p). On consumer routes only Google sells plans: Google AI Pro at $19.99/mo[13] includes a limited Veo 3.1 Lite trial, and full Veo 3.1 starts with Google AI Ultra from $99.99/mo[13]. MiniMax instead offers pay-as-you-go API access plus open H3-Base weights for self-hosting[5].

Snapshot

MiniMax H3 vs Veo 3.1 at a glance (check dates per source in the Evidence list)
SpecMiniMax H3Veo 3.1
StatusGenerally available (public pay-as-you-go API; open-weight H3-Base checkpoints on Hugging Face)[2]Paid preview in the Gemini API (Google's current model list names veo-3.1-generate-preview and veo-3.1-lite-generate-preview as Preview models; Veo 3.1 Fast rates are still published on the pricing page); also served via Vertex AI, the Gemini app and Flow[12]
DeveloperMiniMax[2]Google[12]
Released2026-07-31 (MiniMax H3)[2]2025-10-15[12]
Lowest API price (per second, with audio)from $0.08/s (— 768P)[3]from $0.05/s (Lite)[8]
Cheapest paid consumer planNo public data$19.99/mo Google AI Pro
$99.99/mo Google AI Ultra[13]
Free tierNo public dataGemini API free tier: Not available for Veo 3.1 Standard, Fast or Lite (paid tier only)[8]
Max length, single generation15 s (MiniMax H3)[1]8 s[6]
Max resolution2K[2]4k (8s length only)[6]
Native audioYes (MiniMax H3)[2]Yes[7]
Image-to-videoYes (MiniMax H3)[1]Yes[7]
Public APIYes (MiniMax platform API)[3]Yes[12]

Pricing

Winner: Veo 3.1
Pricing: MiniMax H3 vs Veo 3.1 (check dates per source in the Evidence list)
MetricMiniMax H3Veo 3.1Leader
API price, 720p with audio (per second)$0.08/s (— 768P)[3]$0.40/s Standard
$0.10/s Fast
$0.05/s Lite[8]
Veo 3.1
API price, 1080p with audio (per second)$0.13/s (— 2K)[3]$0.40/s Standard
$0.12/s Fast
$0.08/s Lite[8]
Veo 3.1
API price, 4K (per second)No public data$0.60/s Standard
$0.30/s Fast[8]
—
API billing basisPay-as-you-go per second of output, by resolution: 768P $0.08/s, 2K $0.13/s (H3-Max variant: 480P $0.05/s, 768P $0.08/s). Input audio free; first 5 input images free, then $0.04/image. Prepaid video packages do not support H3.[3]Per second of successfully generated video, with audio included by default; price varies by tier (Standard / Fast / Lite) and by output resolution (720p / 1080p / 4k)[8]—
Cheapest paid consumer planNo public data$19.99/mo Google AI Pro
$99.99/mo Google AI Ultra[13]
—
Consumer plan ladderNo public dataGoogle AI Pro $19.99/month (1,000 Google Flow Credits, limited trial of Veo 3.1 Lite); Google AI Ultra from $99.99/month (10,000 or 25,000 Google Flow Credits, full Veo 3.1)[13]—
In-app cost per secondNo public dataNo public data—
Free tierNo public dataGemini API free tier: Not available for Veo 3.1 Standard, Fast or Lite (paid tier only)[8]—

MiniMax H3 leads 0 rows, Veo 3.1 leads 2, 0 tied; rows marked — have no comparable data or no objective “better”.

Output specs

Winner: MiniMax H3
Output specs: MiniMax H3 vs Veo 3.1 (check dates per source in the Evidence list)
MetricMiniMax H3Veo 3.1Leader
Max length, single generation15 s (MiniMax H3)[1]8 s[6]MiniMax H3
Max length with extensionsNo public data148 s (with extensions, 720p)[6]—
Max output resolutionNo public data2160p (4K) (Standard)[6]—
Frame rate24 fps (MiniMax-H3 (Hugging Face model card))[5]24 fps[6]Tie
Aspect ratios21:9, 16:9, 4:3, 1:1, 3:4, 9:16 and more (API: common ratios or adaptive; input aspect ratio must be within 2:5–5:2)[5]16:9 (default), 9:16[6]—
Max reference inputs per generation12 (— reference generation)[1]3[6]MiniMax H3

MiniMax H3 leads 2 rows, Veo 3.1 leads 0, 1 tied; rows marked — have no comparable data or no objective “better”.

Features

Tie
Features: MiniMax H3 vs Veo 3.1 (check dates per source in the Evidence list)
MetricMiniMax H3Veo 3.1Leader
Native audioYes (MiniMax H3)[2]Yes[7]Tie
Lip-syncNo public dataNo public data—
Text-to-videoYes (MiniMax H3)[1]Yes[7]Tie
Image-to-videoYes (MiniMax H3)[1]Yes[7]Tie
First / last frame controlYes (MiniMax H3)[1]Yes[12]Tie
Multi-shot in one generationYes (MiniMax H3)[2]No public data—
Video editingYes (MiniMax H3)[2]Yes (Lite)[14]Tie
Video extensionNo public dataYes[7]—

MiniMax H3 leads 0 rows, Veo 3.1 leads 0, 5 tied; rows marked — have no comparable data or no objective “better”.

Access

Tie
Access: MiniMax H3 vs Veo 3.1 (check dates per source in the Evidence list)
MetricMiniMax H3Veo 3.1Leader
StatusGenerally available (public pay-as-you-go API; open-weight H3-Base checkpoints on Hugging Face)[2]Paid preview in the Gemini API (Google's current model list names veo-3.1-generate-preview and veo-3.1-lite-generate-preview as Preview models; Veo 3.1 Fast rates are still published on the pricing page); also served via Vertex AI, the Gemini app and Flow[12]—
Public APIYes (MiniMax platform API)[3]Yes[12]Tie
Where to use itMiniMax platform API (platform.minimax.io, pay-as-you-go); open-weight H3-Base checkpoints on Hugging Face (MiniMaxAI/MiniMax-H3) for self-hosting; H3-Max speed variant also priced on the MiniMax platform[5]Gemini API in Google AI Studio and Vertex AI; Gemini app; Flow[12]—
Regional limitsNo public dataIn EU, UK, CH and MENA locations, allow_adult is the only permitted personGeneration value[6]—
Release date2026-07-31 (MiniMax H3)[2]2025-10-15[12]—

MiniMax H3 leads 0 rows, Veo 3.1 leads 0, 1 tied; rows marked — have no comparable data or no objective “better”.

Public leaderboards

Winner: MiniMax H3
Public leaderboards: MiniMax H3 vs Veo 3.1 (check dates per source in the Evidence list)
MetricMiniMax H3Veo 3.1Leader
LMArena text-to-video score1460 ±9 (minimax-h3)[9]1364 ±14 (veo-3.1-audio)[10]MiniMax H3
LMArena text-to-video rank (of 48)#8 (minimax-h3)[9]#12 (veo-3.1-audio)[10]—
LMArena image-to-video score1495 ±5 (minimax-h3)[4]1398 ±10 (veo-3.1-audio)[11]MiniMax H3
LMArena image-to-video rank (of 48)#1 (minimax-h3)[4]#13 (veo-3.1-audio)[11]—
VGI-Bench overall scoreNo public data32.0[15]—

MiniMax H3 leads 2 rows, Veo 3.1 leads 0, 0 tied; rows marked — have no comparable data or no objective “better”.

How rows are judged: a lower price, a longer clip, a higher resolution or frame rate, more reference inputs and a supported feature win. Prices of a discontinued product are shown as historical and cannot win. LMArena scores are human-preference ratings (board snapshot dated 2026-09-21, read 2026-10-07); scores whose ± intervals overlap count as a tie. VGI-Bench is a research benchmark of visual reasoning. A row marked — has no comparable data.

Where each model leads

Where MiniMax H3 leads

Arena standing and input flexibility. H3 holds rank #8[9] on text-to-video and rank #1[4] on image-to-video at LMArena[9][4], and both gaps to Veo 3.1 exceed the published confidence intervals. Reference-to-video takes images, video clips and audio in one request[1], and bracketed camera tags like [pan] and [zoom] steer the virtual camera[1]. MiniMax's earlier video line is covered in our Hailuo 02 review , useful context for the company's trajectory.

Where Veo 3.1 leads

Tier range and finished-video length. Veo 3.1 publishes three audio-included price points per resolution, down to $0.05/s[8] on Lite at 720p[8], supports first-and-last-frame transitions[12], and extends clips to 148 s[6]. It also runs inside the Gemini app and Flow[12], consumer surfaces H3 has no equivalent for[3]. Note that the Gemini API lists no free tier for any Veo 3.1 variant[8].

To price a real brief rather than a rate card, run the same storyboard through AIContentDrop : plans start at $19/mo, or $14/mo billed annually, and the free tier's 10 credits let you test before upgrading[16].

Which should you choose?

Pick MiniMax H3 if…

blind-test ranking is your KPI, you want 15 s[1] single takes, heavy mixed reference input[1], or open weights to run on your own hardware[5].

Pick Veo 3.1 if…

you need 4K or 1080p output[6], videos extended to 148 s[6], a draft tier at $0.05/s[8], or generation inside the Gemini app and Flow[12].

Frequently asked questions

Which ranks higher on LMArena?

MiniMax H3, on both boards: rank #8[9] against rank #12[10] on text-to-video, and rank #1[4] against rank #13[11] on image-to-video. Both score gaps exceed the boards' published confidence intervals.

Which is cheaper per second?

It depends on the tier. Veo 3.1 Lite at $0.05/s[8] (720p) sits below H3's $0.08/s[3] at 768P, and Veo 3.1 Fast at $0.12/s[8] (1080p) is just below H3's 2K rate of $0.13/s[3]. Veo 3.1 Standard, though, costs $0.40/s[8], five times H3's base rate.

Which makes longer videos?

In one pass, H3: 15 s[1] against 8 s[6]. With extensions Veo 3.1 reaches 148 s[6] at 720p; we found no extension feature in MiniMax's H3 documentation.

Can I try either model for free?

Not through the APIs: the Gemini API has no free tier for Veo 3.1[8], and MiniMax publishes no free H3 allowance. H3's open H3-Base checkpoints can be self-hosted at your own compute cost[5].

Does either model output 4K?

Only Veo 3.1, and only on 8-second clips[6], at $0.60/s[8] on the Standard tier. MiniMax states a 2K ceiling for H3[2].

Evidence

Only official sources are cited: vendor pages, and leaderboard publishers for their own rankings. Anything we could not find on such a page is shown as “No public data”.

  1. MiniMax API Docs — Video Generation guide — MiniMax. platform.minimax.io/docs/guides/video-generation. Checked 2026-10-07
  2. MiniMax — 'MiniMax H3' launch announcement — MiniMax. www.minimax.io/blog/minimax-h3. Checked 2026-10-07
  3. MiniMax API Docs — Product Pricing (overview) — MiniMax. platform.minimax.io/docs/pricing/overview. Checked 2026-10-07
  4. Arena (arena.ai) — Image-to-Video Leaderboard — arena.ai. arena.ai/leaderboard/image-to-video. Checked 2026-10-07
  5. Hugging Face — MiniMaxAI/MiniMax-H3 model card (vendor-published) — MiniMax (MiniMaxAI on Hugging Face). huggingface.co/MiniMaxAI/MiniMax-H3. Checked 2026-10-07
  6. Gemini API — Generate videos with Veo — Google. ai.google.dev/gemini-api/docs/veo. Checked 2026-10-06
  7. Gemini API — Video generation docs index — Google. ai.google.dev/gemini-api/docs/video. Checked 2026-10-06
  8. Gemini API — Pricing (Veo 3.1 Standard / Fast / Lite) — Google. ai.google.dev/gemini-api/docs/pricing. Checked 2026-10-06
  9. Arena (arena.ai) — Text-to-Video Leaderboard — arena.ai. arena.ai/leaderboard/text-to-video. Checked 2026-10-07
  10. Text-to-Video Arena leaderboard — LMArena (arena.ai). arena.ai/leaderboard/text-to-video. Checked 2026-10-06
  11. Image-to-Video Arena leaderboard — LMArena (arena.ai). arena.ai/leaderboard/image-to-video. Checked 2026-10-06
  12. Introducing Veo 3.1 and new creative capabilities in the Gemini API — Google Developers Blog. developers.googleblog.com/introducing-veo-3-1-and-new-creative-capa…. Checked 2026-10-06
  13. Google AI plans (Google One, US pricing) — Google. one.google.com/intl/en_us/about/google-ai-plans. Checked 2026-10-06
  14. Gemini API — Models list (Veo 3.1, Veo 3.1 Lite) — Google. ai.google.dev/gemini-api/docs/models. Checked 2026-10-06
  15. VGI-Bench leaderboard (EMNLP 2026) — VGI-Bench research team. hexuan21.github.io/VGI-Bench. Checked 2026-10-06
  16. AIContentDrop — Pricing — AIContentDrop. aicontentdrop.com/pricing. Checked 2026-10-07