MiniMax H3 Max vs MiniMax H3 vs H3 Max Turbo

MiniMax H3 Max vs H3 on Makernice: MiniMax H3 goes up to 2K and 4K (upscaled from 768p) with 5 reference images, from $0.06 per second. H3 Max tops out at 1080p with 4 references and no extra options, from $0.0452. H3 Max Turbo starts at $0.03 per second but has no reference-to-video mode.

What's the difference between MiniMax H3, H3 Max and H3 Max Turbo?

Source: GET /v1/models and GET /v1/pricing; native/upscaled and audio notes from the Makernice catalog registry. Catalog facts only: we haven't compared outputs, so this post makes no quality ranking.

How much does each cost per clip?

Same price in every mode a model supports. prompt_expansion doesn't change the price. Clip price = per-second price × seconds, rounded to the cent; the quote shows the exact amount.

Billed in exact micro-USD; a sum of rounded clip prices can differ by a cent. Live: Pricing or GET https://api.makernice.com/v1/pricing (no key needed).

Against MiniMax's own list prices (the official_unit_price field of GET /v1/pricing), it goes both ways:

MiniMax's pricing page doesn't list H3 at 480p or 4K, H3 Max at 1080p, or H3 Max Turbo, so those rows have no official comparison. Official list prices come from GET /v1/pricing; they change often.

Quote all three models · or price a batch in the cost calculator

Is H3 Max faster than H3?

MiniMax's API reference calls MiniMax-H3-Max "the fast generation variant" (platform.minimax.io). That is MiniMax's description; we haven't published our own render times on Makernice, so this post makes no speed claim for H3 Max or H3 Max Turbo. To check on your own workload, send the same prompt to each model id and time each job from create to succeeded.

When does MiniMax H3 make sense?

• 2K or 4K delivery. H3 is the only one of the three with 2K and 4K tiers. Both are upscaled from 768p, so treat them as a delivery size, not extra native detail.
• Up to 5 reference images in reference-to-video mode.
• Prompt expansion. prompt_expansion (off, balanced or quality) is an option on the same per-second price, with no separate surcharge on Makernice's price list.

When are H3 Max or H3 Max Turbo enough?

• 768p or 1080p output. Both offer 480p, 768p and 1080p; H3 Max's 1080p is refined from 768p.
• Need references: H3 Max takes up to 4 reference images.
• Drafts and non-reference work: H3 Max Turbo covers text-to-video, image-to-video and first/last frame at $0.24 per 5-second 768p clip. It has no reference-to-video mode.

How do you switch between them in one request?

Change model and keep the rest, with three adjustments:

• To H3 Max: remove seed and prompt_expansion (H3 Max takes no options), and use 480p, 768p or 1080p.
• To H3 Max Turbo: don't use r2v or reference_images; use 480p, 768p or 1080p.
• H3 Max 1080p jobs: switch to H3 at 768p, 2k or 4k.

Fields a model doesn't support return 400 invalid_request before anything is held, so quote first.

import os, requests

API = "https://api.makernice.com/v1"
H = {"Authorization": f"Bearer {os.environ['MAKERNICE_API_KEY']}"}
base = {"mode": "t2v", "resolution": "768p", "duration": 10, "aspect_ratio": "16:9",
        "prompt": "A cyclist crosses a rain-soaked bridge at night, neon reflections, tracking shot."}

jobs = {
    "minimax-h3":           {**base, "seed": 7, "prompt_expansion": "balanced"},
    "minimax-h3-max":       base,                                   # no options on H3 Max
    "minimax-h3-max-turbo": {**base, "seed": 7, "prompt_expansion": "balanced"},
}
for model, job in jobs.items():
    q = requests.post(f"{API}/generations/quote", json={**job, "model": model}, headers=H).json()
    print(model, f"${int(q['amount']) / 1e6:.2f}")   # $0.72 · $0.69 · $0.48

Quoting is free and holds nothing; each quote is locked for 12 minutes. Create the generation with the quote_id and an Idempotency-Key, then poll (quickstart).

Where does each one win?

• MiniMax H3: the only one of the three with 2K and 4K and with 5 references, but those tiers are upscaled from 768p. At 2K it costs more than MiniMax's own list price ($0.156 vs $0.13 per second) and more than WaveSpeed (USD 0.14) and Atlas Cloud (USD 0.13; both checked October 22, 2026).
• MiniMax H3 Max: up to 4 references and a 1080p tier; below MiniMax's list price at 480p and 768p.
• MiniMax H3 Max Turbo: priced below H3 and H3 Max at every resolution they share, from $0.03 per second, but no reference-to-video mode.
• Elsewhere: MiniMax's own platform lists H3 at 2K for $0.13 per second.
• Output quality: we haven't compared outputs, so we don't rank them.

Quote all three models · or price a batch in the cost calculator

Related

• Models: MiniMax H3 · MiniMax H3 Max · MiniMax H3 Max Turbo · All 16 models
• Comparisons: Wan 3.0 vs MiniMax H3 · MiniMax H3 vs Seedance 2.5
• Guides: Duration & resolution limits · Reference-to-video guide
• Cost: AI video cost calculator · No charge for failed generations · Pricing

About Makernice

Makernice is an AI video and image generation API: one API key and one prepaid balance for 16 models from ByteDance, Kuaishou, MiniMax and Alibaba. Every run starts with a quote that locks the price for 12 minutes, through Makernice's own async REST API (quote, create, poll, cancel).