Free weights, paid hardware

Free weights are not zero cost: you buy the GPU up front (one-time, estimated — prices may rise), then pay for power, cooling and your own time. Six of the twelve are Apache 2.0 (HappyHorse-1.0, SANA-WM, Bernini, MAGI-2, LingBot-Video, LongLive 2.0); LTX-2 and LTX-2.5 use community licences, SANA-Video uses the NVIDIA Open Model License, and the rest ship custom licences. Nine were open-sourced in 2026; the other three are overseas 2025–2026 models (LTX-2, SANA-Video, LongLive 2.0). The hardware ladder still runs from an 18.5GB peak on a single RTX 5090 to 8× H100 — a discrete GPU is the floor, not an option.

Apache 2.0, commercially usable

¥24,000+

One-time hardware · estimate, may rise

models · 9 in 2026 + 3 overseas 2025–2026

Online models 20 Model API · 13 Platforms · 7 Open source · 12 Overseas · 7

Online models · 20Model APIs and platform workbenches, cheapest first; the 20 online quotes contain no estimates. The 12 open-source models have no per-second price, so their one-time GPU cost (an estimate that may rise) is shown on a separate page.

Range ¥0.09 – ¥5.05Spread 56×

Model layer · 11Official API keys billed per second. You assemble the workflow yourself.

Range ¥0.09 – ¥5.05Quotes 11

Platform layer · 6Credits converted to cost per second. Switching model inside one platform can change the price 4×.

Range ¥0.476 – ¥2.28Quotes 7

Open source · 12Free weights are not free hardware: you buy the GPU up front. One-time estimates, subject to price rises. Six of the twelve are Apache 2.0; the floor is a single 5090, the ceiling 8× H100.

Apache 2.0 6 / 12Hardware from ≈ ¥24,000

Overseas · 7Official per-second API prices from Google / Runway / Luma / xAI / Pika, converted at the 2026-11-10 PBoC mid rate (1 USD = ¥6.7167) so they sort in the same unit as the domestic rows; the USD list price stays in each subtitle. Subscription-only tools (Midjourney, Firefly) and the retired Sora 2 API are not included.

Range ¥0.134 – ¥0.806Spread 6×

Open source Highest bar

Weights are free, hardware is not: you buy the GPU up front (one-time estimate, may rise) and still pay for power and time. Six of the twelve are Apache 2.0 (HappyHorse-1.0, SANA-WM, Bernini, MAGI-2, LingBot-Video, LongLive 2.0); LTX-2 and LTX-2.5 use community licences and SANA-Video uses the NVIDIA Open Model License. Hardware runs from one RTX 5090 to 8× H100. See all 12 →

Vendor

First 9 sorted by VRAM floor; last 3 are overseas open-source baselines · one-time prices are estimates Specs checked 2026-11-10 · hardware 2026-10-21 · 0 days ago

Price comparison table
#	Platform / model	Layer	VRAM · setup	One-time hardware
—	OSSMatrix-Game 4.0 ↗Skywork · 2026-10 · code + weights + white-box · Wan2.2 5B base	Open source	18.5 GBMeasured peak · one 5090, real time	≈ ¥35,000 · one-time1× RTX 5090 32GB · estimate, may rise
—	OSSLTX-2.5 ↗Lightricks · 2026-08 · community licence · 22B + 12B encoder	Open source	24 GB12GB only with FP8 + offload	≈ ¥24,000 · one-time1× RTX 4090 24GB · estimate, may rise
—	OSSSkyReels-V3 ↗Skywork · 2026-01 · custom licence · 14B / 19B at 720P	Open source	24 GBUnder 24GB needs FP8 + offload	≈ ¥24,000 · one-time1× RTX 4090 24GB · estimate, may rise
—	OSSSANA-WM ↗NVIDIA · 2026-05 · Apache 2.0 · 2.6B	Open source	32 GBDistilled + NVFP4 · one 5090	≈ ¥35,000 · one-time1× RTX 5090 32GB · estimate, may rise
—	OSSHappyHorse-1.0 ↗Alibaba ATH · 2026-04 · Apache 2.0 · 15B	Open source	32 GB30GB+ in FP16 · 24GB needs quant	≈ ¥35,000 · one-time1× RTX 5090 32GB · estimate, may rise
—	OSSBernini ↗ByteDance · 2026-06 · Apache 2.0 · 14B	Open source	80 GBOfficial: H100/H800 (FA3)	≈ ¥250,000 · one-time1× H100 80GB · estimate, may rise
—	OSSLingBot-Video ↗Ant Lingbo · 2026-07 · Apache 2.0 · 30B-A3B MoE	Open source	80 GB84GB peak per card, 38GB with FSDP	≈ ¥2,000,000 · one-time8× H100 80GB · estimate, may rise · cloud rental is more practical
—	OSSMiniMax H3-Base ↗MiniMax · 2026-08 · community licence · 33B + 32B encoder	Open source	4× 80 GBOfficial example: 4× 80GB	≈ ¥1,000,000 · one-time4× H100 80GB · estimate, may rise · cloud rental is more practical
—	OSSMAGI-2 Preview ↗Sand.ai · 2026-08 · Apache 2.0 · 114B MoE	Open source	8× 80 GB8× H100 · 228GB BF16 weights	≈ ¥2,000,000 · one-time8× H100 80GB · estimate, may rise · cloud rental is more practical
—	OSSLTX-2 ↗Lightricks (IL) · 2025-10 · community licence · 19B	Open source	24 GB38GB BF16; FP8/FP4 fits 24GB	≈ ¥24,000 · one-time1× RTX 4090 24GB · estimate, may rise
—	OSSSANA-Video ↗NVIDIA + MIT (US) · 2025-10 · NVIDIA Open Model License · linear DiT	Open source	32 GBRTX 5090 · NVFP4 · 5s 720P	≈ ¥35,000 · one-time1× RTX 5090 32GB · estimate, may rise
—	OSSLongLive 2.0 ↗NVIDIA (US) · 2026-05 · Apache 2.0 · Wan2.2 5B + NVFP4	Open source	1× GB200Test rig 180GB · NVFP4 45.7 FPS	Cloud only1× GB200 · not sold standalone

Nothing matched — try another keyword.

Methodology: how these numbers are computed

CNY/sec = official list price ÷ billed seconds; credit platforms = plan price ÷ seconds that tier can produce.
Overseas layer = USD list price × PBoC (CFETS) mid rate (currently 2026-11-10, 1 USD = ¥6.7167).
Open source has no per-second price — the cost is one-time hardware (GPU estimates, may rise) plus power and time, hence a separate page.
Subscription-only tools (Midjourney, Firefly) and retired APIs are not collected; anything computable is computed. The only estimates are the open-source GPU prices, labelled as such and subject to change — check live listings before buying.

Three traps worth knowing. ① Official retail is not the cheapest — Seedance 2.5 at ¥39 / 15 sec (¥2.6/sec) costs more than most platform annual plans. Platforms subsidise with annual prepay to lock you in. ② Open source ≠ commercially usable — six of the twelve are Apache 2.0, but LTX-2 / LTX-2.5 and MiniMax H3 use community licences (H3 needs a separate application in the US/EU/UK/KR), and SANA-Video uses the NVIDIA Open Model License, so read the terms before you ship. ③ Open source ≠ free — the weights cost nothing, but the GPU is a one-time purchase (estimated here, and prices can rise), plus power and your time.

⚖

Data on this page is indicative only — listing a vendor is not a recommendation or guarantee.Figures are compiled and converted from public pages and can change at any time; check the vendor's current pricing before you buy. Read the disclaimer