MiniMax H3 API Pricing: Every Provider Compared

MiniMax H3 has one of the widest hosting markets of any video model we track — eleven separate providers, including MiniMax's own direct API. That much competition usually compresses price at the bottom while leaving at least a couple of hosts pricing near the model creator's own rate, and H3 follows that pattern closely.

Full pricing table, 768p tier

Provider	Price (768p)
MachGen	$0.0400/sec
TOAPIS	$0.0600/sec
Fal	$0.0600/sec
Novita	$0.0690/sec
Atlas Cloud	$0.0800/sec
MiniMax (direct)	$0.0800/sec
Replicate	$0.0800/sec
Pika	$0.0800/sec
WaveSpeedAI (native)	$0.0800/sec
WaveSpeedAI-resell	$0.1000/sec

Five separate hosts — Atlas Cloud, MiniMax direct, Replicate, Pika, and WaveSpeedAI's native listing — all land on exactly $0.08/sec, which functions as the de facto market rate for H3 at this tier. MachGen undercuts that rate by 50%, and Novita and Fal/TOAPIS both sit meaningfully below it too. The one host actually pricing above market is WaveSpeedAI's own resale listing (distinct from its native $0.08 listing) — a case where the same host carries two different listings for the same model at two different prices, which is worth checking for specifically since it's easy to accidentally integrate against the pricier of the two.

MachGen's consistent pattern

This is the third model in this series (alongside Seedance and Wan 3.0) where MachGen comes out as the cheapest or among the cheapest hosts. That's a strong enough pattern across independent models that it's worth treating as a real signal about MachGen's general pricing strategy — thin margins across its catalog — rather than a coincidence specific to any one model. It doesn't mean MachGen is automatically the right host for every model (check each one specifically; hosting markets vary), but it's a reasonable first place to check.

Other tiers

H3 also ships 480p and 1440p/2K tiers at several hosts:

Provider	480p	1440p / 2K
MachGen	$0.035	$0.100
Fal	$0.050	$0.130
Atlas Cloud	$0.038	$0.080-$0.115
MiniMax (direct)	—	$0.130
Novita	—	$0.110

Atlas Cloud's 1440p pricing varies by whether you're requesting the standard or enhanced-super-resolution variant of that tier — check the specific tier string against Atlas Cloud's own listing before assuming a single flat 1440p rate.

What this costs at real volume

A product generating 4,000 five-second H3 clips a month at 768p:

• MachGen: 4,000 × 5 × $0.04 = $800/month
• Market-rate cluster (Atlas Cloud, MiniMax, Replicate, Pika, WaveSpeedAI native): 4,000 × 5 × $0.08 = $1,600/month
• WaveSpeedAI-resell: 4,000 × 5 × $0.10 = $2,000/month

The gap between the cheapest and most expensive listing here is $1,200/month at this volume — and importantly, that gap exists between two listings on the same host (WaveSpeedAI native vs. WaveSpeedAI-resell), not just between different companies. Always verify which specific listing you're integrated against.

A second worked example: comparing tiers, not just hosts

The tier choice matters as much as the host choice for H3. At MachGen, moving from 480p to 768p to 1440p roughly triples the rate at each step ($0.035 → $0.04 → $0.10). A team generating 6,000 five-second clips a month:

• 480p, MachGen: 6,000 × 5 × $0.035 = $1,050/month
• 768p, MachGen: 6,000 × 5 × $0.04 = $1,200/month
• 1440p, MachGen: 6,000 × 5 × $0.10 = $3,000/month

The jump from 768p to 1440p costs 2.5x more than staying at 768p, while the jump from 480p to 768p costs barely more (14%) at the same host. That asymmetry means 768p is close to a "free" upgrade over 480p at MachGen specifically, while 1440p is a real, deliberate spend decision — worth checking per host, since the tier-to-tier ratio isn't necessarily identical everywhere in the table.

Frequently asked questions

Why does WaveSpeedAI carry two separate H3 listings at different prices? This typically happens when a host adds a new, better-integrated listing for a model (WaveSpeedAI's "native" listing) alongside an older resale-style listing it hasn't retired — the native listing is usually the more current one to prefer, and in this case it's also the cheaper of the two.

Is MiniMax's own direct rate ever competitive? At 768p, MiniMax's direct rate ($0.08/sec) ties with four other hosts rather than beating any of them — consistent with the pattern seen with Alibaba's Wan 3.0 and Google's Veo 3.1 in this series, where the model creator's own listing is rarely the cheapest option.

How many hosts is "too many" to reasonably evaluate before shipping? There's no fixed number — the practical approach is checking the top 2-3 cheapest verified hosts for your specific tier and volume, rather than exhaustively benchmarking all eleven before you can ship anything. VideoRouter's own routing already biases toward the cheaper end of the roster by default, which reduces how much manual evaluation you need to do upfront.

Does H3's pricing pattern generalize to other MiniMax models? Not necessarily — check each model's own listing. H3 having eleven hosts is a specific fact about H3's licensing and adoption, not a property of every MiniMax model.

Is Novita's H3 pricing related to its Kling pricing? No direct relationship — Novita hosts multiple model creators' checkpoints independently, and its pricing strategy on one model (see Kling v3.0 API Pricing, where Novita ties WaveSpeedAI exactly) doesn't predict its pricing on another. Check each model separately.

How to call it

import requests

job = requests.post(
    "https://videorouter.sh/api/v1/videos",
    headers={"Authorization": "Bearer llmr_sk_live_...", "Content-Type": "application/json"},
    json={
        "model": "minimax/h3",
        "prompt": "a paper airplane gliding over a city",
        "duration_secs": 5,
    },
).json()

Pinning to a specific host, MachGen included, is a provider field away — see How to Call the MiniMax H3 API for a full walkthrough with provider pinning, and /docs/video-generation for the complete request reference. VideoRouter's MiniMax H3 model page keeps this pricing comparison live.