LTX 2.5 vs MiniMax H3: Which Open Model to Pick?

Most pages ranking for these two models were built to sell one of them. Search for either name and you land on a single-model site or a vendor's own "alternatives" page. Each tells half the story, and the half it leaves out is usually the half that matters to you.

The usual shortcut, "LTX is open, H3 is closed", is also wrong. Both are open-weight models released in summer 2026: MiniMax announced H3 on July 31, and Lightricks shipped LTX-2.5 on August 11. But they are open in very different ways. For a developer, the LTX 2.5 vs MiniMax H3 decision can come down to where your company is incorporated before it ever gets to video quality.

This comparison is based on Lightricks' LTX-2.5 API docs, release notes and license summary, MiniMax's H3 announcement, API docs, model card and license, and on how Vogoo runs both families in production: LTX 2.5 Pro and Fast, MiniMax H3 and the speed-tuned H3 Max. We haven't run a formal benchmark between them, so you won't see invented quality scores. You will get the licensing facts, the speed tiers, the resolution limits and the credit cost of the same clip on each.

By the end you'll know which one to self-host (if either), which one to call through a studio, and which tier to use for drafts vs finals.

LTX 2.5 vs MiniMax H3 at a glance

You can start an LTX 2.5 clip or open the MiniMax H3 generator side by side while you read.

Open weights, two very different licenses

This is the dimension one-sided pages skip, and for developers it's often decisive.

LTX 2.5. Lightricks publishes the full LTX-2.5 weights on Hugging Face under the LTX-2.x Community License. Per its model card, commercial and production use is free for entities under $10M in annual revenue, measured across the whole entity including affiliates. Above that you need a paid commercial agreement. Lightricks also notes that transferring fine-tunes may require a paid license. The weights ship in a component-split pack (distilled and full transformers, text encoder, video and audio VAEs, upscalers), plus a raw pretrained checkpoint meant for deep fine-tuning.

MiniMax H3. MiniMax publishes the H3 base weights on Hugging Face under the MiniMax H3 Community License. Three clauses matter:

1. Territory. The license applies worldwide excluding the European Union, the United Kingdom, South Korea and the United States. Organizations in those regions can apply to MiniMax for a separate license. MiniMax's license Q&A says the API remains available globally.
2. Attribution. Commercial products that use H3 must prominently display "MiniMax H3" in the user interface.
3. Scale. Commercial products above $20 million in yearly revenue need prior written authorization.

There's also a scope difference. The open H3 release covers H3-Base, which renders at 768p. The H3-Context-IR prompt processor and the H3-Regenerate-2K module are not open-sourced; MiniMax offers them through its API. So "self-hosted H3" and "H3 at 2K" are not the same thing today. H3 Max's weights aren't published at all.

Rule of thumb: if self-hosting or fine-tuning is part of your plan, read the license before the benchmark. For a US or EU team, LTX 2.5 is the only one of the two you can download and ship without first negotiating a license.

Speed tiers: LTX Fast vs Pro vs H3 vs H3 Max

Both vendors split their lineup into a speed tier and a quality tier. The naming just doesn't line up.

• LTX 2.5 Fast is described by Lightricks as the variant "for speed and low cost". In the API it's the one that runs long: up to 20 seconds at 720p or 1080p.
• LTX 2.5 Pro is the "higher fidelity" variant, with 6, 8 or 10 second clips.
• MiniMax H3 is the full model, and the only one of the four with a 2K tier.
• MiniMax H3 Max is described in MiniMax's docs as post-trained on H3 and optimized for high-speed generation. It outputs 480P or 768P and generates faster than H3.

On raw speed, both vendors publish their own charts. Lightricks' LTX-2.5 page, for example, ranks LTX ahead of H3 on generation time. Treat any vendor chart as a vendor claim: the hardware, queue and resolution differ from row to row. The only speed number that matters is your own wait time for the resolution you'll deliver.

Resolution and duration: 4K vs 2K, 20s vs 15s

Resolution. LTX 2.5 reaches 4K (3840×2160 landscape) in the API on both Fast and Pro. H3 reaches 2K, and it gets there differently from an upscaler. MiniMax's model card describes H3-Regenerate-2K as feeding the 768p result and the original prompt context back into the model to regenerate it at 2K, so fine detail like small text is re-derived rather than guessed. LTX 2.5 has its own answer to fidelity: Lightricks' release notes describe a new diffusion video decoder for sharper faces and legible text, and "Diffusion Fidelity Rendering", which allocates compute by scene complexity.

Duration. LTX 2.5 Fast runs up to 20 seconds at 720p/1080p. H3 runs up to 15. LTX also has an optional auto duration mode that predicts clip length from the prompt. Lightricks' docs note that on prepaid API accounts, credits are held against the longest possible length until the job finishes.

On Vogoo, LTX 2.5 Pro runs at 720p or 1080p for 6, 8 or 10 seconds. LTX 2.5 Fast runs from 720p up to 4K for 6–20 seconds, in 16:9 or 9:16. MiniMax H3 runs at 2K and H3 Max at 768P, both 5–15 seconds in six aspect ratios (21:9 to 9:16). One difference to know before you pick: LTX 2.5 includes native audio and image-to-video on Vogoo, while H3 and H3 Max currently run text-to-video only, without listed audio.

LTX 2.5 vs 2.3: what changed

If you already use LTX-2.3 (released March 26, 2026, with a 22B model, native audio and up to 4K at 50 fps), Lightricks' release notes list what 2.5 adds:

• native multi-shot scenes that hold character, lighting and voice across cuts
• a diffusion video decoder
• a custom Gemma 4 12B text encoder
• a prompt enhancer
• auto duration
• a substantially better distilled model
• a raw pretrained checkpoint for fine-tuning

The headline change is multi-shot. LTX's earlier versions produced one continuous shot per generation.

What other comparisons miss: the credit cost of the same clip

"ltx 2.5 vs h3" pages argue about quality without pricing the same deliverable. Here is one 10-second clip across every tier Vogoo runs, at list price:

Three things stand out:

• The whole field sits in a narrow band. Apart from 4K, every option lands between 270 and 660 credits for 10 seconds. At these prices, retries and fit matter more than the rate card.
• H3 at 2K costs about the same as LTX Pro at 720p (430 vs 420 credits). If output resolution is the spec, H3 is the efficient pick. If you need sound or an image-led first frame on Vogoo, LTX gives you both at that price.
• H3 Max is the cheapest draft tier at 270 credits, and LTX Fast 720p is close behind at 320 with audio included.

Cost on Vogoo (as of September 30, 2026)

Vogoo charges per second of output, and the studio shows the exact cost before you generate.

• LTX 2.5 Pro: 42 credits per second at 720p, 59 at 1080p. 6, 8 or 10 seconds. Native audio included.
• LTX 2.5 Fast: 32 credits per second at 720p, 45 at 1080p, 66 at 1440p, 104 at 4K. 6–20 seconds. Native audio included.
• MiniMax H3: 43 credits per second at 2K. 5–15 seconds.
• MiniMax H3 Max: 27 credits per second at 768P. 5–15 seconds.

Worked example: a 20-second 1080p explainer shot. LTX 2.5 Fast does it in one generation: 20 × 45 = 900 credits. On H3, the longest single generation is 15 seconds (645 credits at 2K), so the same 20 seconds takes two generations and a cut. You'd spend 860 credits for 20 seconds (430 + 430), plus the risk that the two halves don't match.

When a draft is all you need, run a quick H3 Max draft at 768P before you commit to a final render.

LTX 2.5 vs MiniMax H3: which should you choose?

This is the "minimax vs ltx" decision in one list.

Choose LTX 2.5 if:

• You want to self-host or fine-tune, and you're in the US, EU, UK or South Korea, or you can't show third-party branding.
• You need 4K, 48/50 fps, clips longer than 15 seconds, or multi-shot scenes.
• You want audio and image-to-video in the same generation on Vogoo.

Choose MiniMax H3 if:

• You want 2K from a text prompt at the lowest per-second price for that resolution.
• The shot depends on crisp detail, such as a product label or on-screen type.
• You'll use MiniMax's own API for its reference modes (up to 9 images, 3 videos, 3 audio clips).

Choose H3 Max or LTX Fast 720p if:

• You're iterating on composition and want the cheapest reasonable draft.

Rule of thumb: creators should pick by deliverable (4K, long or with audio means LTX; sharp 2K text-led means H3). Developers should pick by license first, because the territory clause settles it for many teams before quality comes up.

If Seedance 2.5 is also on your list, see our Seedance 2.5 vs MiniMax H3 comparison.

Run the same prompt on both

1. Write one text-only prompt. For example: "A neon ramen shop sign flickers on in the rain, a cyclist passes, slow dolly left, reflections on wet asphalt." It tests fine text (the sign), motion and camera following at once.
2. Match the settings. 10 seconds, 16:9. Run LTX 2.5 Pro at 1080p (590 credits), MiniMax H3 at 2K (430) and H3 Max at 768P (270).
3. Compare. Check sign legibility, the cyclist's motion, whether the dolly direction was followed, and whether LTX's generated audio saves you an edit.
4. Total spend: 1,290 credits for a three-way answer on your own style of prompt.

Troubleshooting

• Symptom: the LTX clip has the right look but the action lands too early or too late. Cause: the fixed duration doesn't fit the described action. Fix: shorten the prompt to one action, or try a different length (Pro accepts 6, 8 or 10 seconds).
• Symptom: the H3 text is sharp but the scene looks static. Cause: no explicit motion verb or camera instruction. Fix: name one physical action and one camera move, in that order.

FAQ

Is LTX 2.5 better than H3?

Neither wins everywhere. LTX 2.5 goes further on resolution (4K), length (20s), frame rate and licensing freedom. H3 offers 2K at a lower per-second price than LTX's 1080p Pro and a large multimodal reference system in MiniMax's API. Run the same prompt on both for your use case.

Is MiniMax H3 open source like LTX?

Both publish weights, but the terms differ. H3's community license excludes the US, EU, UK and South Korea, requires "MiniMax H3" branding in commercial UIs, and the open release renders at 768p. LTX 2.5's license is free for entities under $10M in revenue, with no territory exclusion in the terms we reviewed.

LTX 2.5 vs MiniMax: which is cheaper per clip on Vogoo?

It depends on the tier. H3 Max at 768P is cheapest (270 credits for 10 seconds). LTX 2.5 Fast 720p is next at 320, with audio. At the top end, H3 2K (430) costs less than LTX Pro 1080p (590) or LTX Fast 4K (1,040).

What do Reddit users say about LTX 2.5 vs MiniMax H3?

Forum threads are useful for seeing other people's prompts and local setups, but they're anecdotal and hardware-dependent, so we don't treat them as evidence. The licensing facts are in each vendor's license file. For quality, the same-prompt test above gives you an answer for your own work.

Can I run MiniMax H3 at 2K on my own GPU?

Not with the open release alone. MiniMax's model card says the open H3-Base renders at 768p. The 2K regeneration module and the Context-IR prompt processor are not open-sourced, and MiniMax offers them through its API.

Bottom line

For LTX 2.5 vs MiniMax H3, the license and the deliverable decide more than the benchmarks do. LTX 2.5 is the safer open-weights bet for most US and EU developers, and the better studio pick for 4K, longer clips and built-in audio. H3 is the efficient choice for sharp 2K text-to-video. Settle it with one prompt: generate it with LTX 2.5 on Vogoo, then run the same prompt on H3 and H3 Max and compare the three results.

Keep reading

• Seedance 2.5 vs MiniMax H3: resolution vs duration, priced per clip
• Kling 3 vs Seedance 2.5 for narrative video
• Driving a video from an audio track with Seedance 2.5

Written by the Vogoo team, September 30, 2026, from Lightricks' LTX-2.5 API documentation, release notes and Hugging Face model card, MiniMax's H3 announcement, API documentation, model card and community license, and Vogoo's own studio configuration. Model specs and credit prices change — check the linked sources and the Vogoo studio before committing a production workflow.

Sources

• LTX-2.5 supported models — LTX API Documentation — Fast vs Pro resolutions, fps, durations, auto duration and prepaid billing hold
• Lightricks/LTX-2.5 model card — Hugging Face — open weights, LTX-2.x Community License, $10M revenue threshold, fine-tune transfer note
• LTX release notes — Lightricks — August 11, 2026 LTX-2.5 release and features; March 26, 2026 LTX-2.3 specs
• MiniMaxAI/MiniMax-H3 model card — Hugging Face (official MiniMax org) — 33B model, 768p open base, API-only 2K module, 24 fps, 32 kHz stereo, license and territory Q&A
• Video Generation guide — MiniMax API Docs — H3 and H3 Max resolutions, durations, input limits, H3 Max description