LongCat-Video
Meituan · video gen · reviewed by Stan K. · Source: data snapshot · data as of 2026-11-12
A clean MIT license from Meituan at 13.6B dense — and no verified hardware figure published yet.
#11 of 14 in Video Gen
ToolScore breakdown — why 44/100
10 criteria × 10 points. Low pillars are visible on purpose — that's the point of a score you can audit. Methodology →
What it does well
- ▸MIT license from Meituan — clean terms, no caps or territories
- ▸13.6B dense: a serious parameter count under a permissive license
- ▸Electricity economics on whatever hardware it turns out to need
Where it costs you
- ▸No verified hardware figure published — verify pending
- ▸Nothing about practical running conditions can be sourced yet
- ▸Dense at 13.6B means the hardware bar is probably not low
Hidden traps — the stuff the pricing page doesn't say
The VRAM cell is '—': no verified hardware figure has been published for LongCat-Video. Any number you see elsewhere is community anecdote until a source confirms it — including ours.
Know the limits before you pay: read the traps above, then decide with eyes open.
Affiliate link where available — never affects the score.
- internal ToolScore Video Gen hub — snapshot 2026-11-12 (numbers from LICENSE files and model cards) verified
Data as of 2026-11-12. Prices and limits change monthly — report anything stale. How we verify: methodology.
Closest alternatives
5-second clips with audio under MIT — but the weights are gated, so "open" here means ask first.
The big Apache-licensed sibling with an official 80 GB target — not a consumer card; the 3060 path e
Apache 2.0 weights whose README says ~60 GB while the ComfyUI path runs under 20 GB on one card — pl
LongCat-Video review
In-depth analysis · last updated 2026-11-12
A clean license, an empty spec sheet
LongCat-Video is MIT from Meituan, 13.6B dense — on paper, one of the most permissive licenses attached to one of the bigger parameter counts in the local table. In practice, our snapshot has no verified hardware figure for it: the VRAM cell stays '—', verify pending. A clean license on a model you cannot size is a promise, not a product.
What MIT buys you
The terms are the story here: MIT means no revenue cap, no territory list, no gated weights — the same class of freedom as Apache, from a major vendor. If the hardware figure arrives and lands consumer-side, this row moves up the table on license strength alone. Until then, the honest position is: free to run, unknown to plan.
How the pillars land
Price/Value 3.5 — free license, unknown entry cost; value cannot be verified until the hardware figure is. Length & resolution caps 2.5 — no caps or hardware figures published. Render speed 5 — neutral: no render-time figure exists. Training use & license 9.5 — MIT, clean. API access 7 — the weights exist under MIT; with no verified hardware figure there is no confirmed runnable path. Feature set 5 — neutral: no capability claims beyond the parameter count appear in our snapshot. Track record 2 — nothing about this model's real-world behavior is verified yet. Ease of use 2.5 — you cannot even plan the rig. Pricing transparency 2 — the absence of published numbers is the whole story of this row. Support & docs 5 — neutral: no support-channel facts in our snapshot.
Verdict — 44/100, UNVERIFIED
Watch this row. The license is right, the size is right, and the moment a verified hardware figure lands it earns a real score. Today it is a clean MIT license wrapped in a question mark.