A game artist pastes a sketch of a lantern into a box, picks "symmetry on," and walks away with a clean, riggable 3D model in the time it takes to pour coffee. The file is not a pretty ball that needs rebuilding; it is something a team can drop into an engine. That is the bar VAST says it has crossed.
Key takeaways
- VAST (三启万物), a Beijing 3D large-model (通用大模型) company founded in 2023 by Song Yachen (宋亚宸), released Tripo 3.0 in August 2025.
- Tripo 3.0 runs at about 20 billion parameters (200亿), roughly ten times its prior generation, and VAST says it generates a 3D model in under ten seconds.
- The model adds pose control, symmetry control, multi-view image input, and smooth-detail control, and can auto-bind a skeleton from a T-pose model for animation.
- VAST reports more than 3 million developers and tens of thousands of enterprise customers using Tripo, with over 40 million 3D models generated.
- VAST's B and B+ rounds, announced together on September 1, 2026, totaled about ¥3 billion (≈ US$420M / HK$3.3B) per Xinhua Finance and Beijing E-Town disclosures, and its tools are integrated with Unity, Blender, Tencent, and NetEase workflows.
From "2D to 3D" to "3D native"
Most early 3D generators took the easy path: spin up a model from 2D images. The problem is that 2D-to-3D loses information that can never be recovered — the back of an object, its inside, its physical consistency. VAST bet on the harder route, training directly in 3D space so the output carries structure, not just a convincing shell.
Tripo 3.0 is the public result of that bet. At 20 billion parameters it is an order of magnitude larger than VAST's earlier versions, and the company says the jump shows up as cleaner geometry and finer texture rather than just more varied output. The practical features are the ones artists actually argue about: symmetry control for anything that should mirror itself, pose control for characters, and multi-view input so a few photos can pin down a shape instead of one ambiguous angle.
Why "game-ready" is the real claim
A 3D model that looks right in a preview is cheap to make now. A model that survives a production pipeline is the hard part. VAST's pitch is "production-ready": the asset should arrive with the topology, bindings, and material slots a studio expects, so the time saved is real instead of borrowed and repaid in cleanup.
The skeleton auto-binding is the clearest example. Feed Tripo a T-pose character and it returns a rigged model, the step that normally sits between "I have a mesh" and "I can animate it." Combined with a smart texture brush for materials, the loop from idea to placeable object shrinks from hours or days to seconds. Whether every output truly needs no fix-up is something only a studio's own pipeline can confirm, but the direction — fewer manual reconstruction steps — is where the value sits.
Where it is already running
VAST reports the model inside real workflows rather than demos. On the games side, it cites integration with Tencent and NetEase titles, including the open-world game Where Winds Meet (燕云十六声) and the social game Eggy Party (蛋仔派对), where user-uploaded photos become interactive 3D props. On the tooling side, it connects to Unity and Blender, the standards most studios already use. On hardware, it names Bambu Lab (拓竹) and Creality (创想三维) among 3D-printing partners, where a 2D product photo becomes a printable model in minutes.
The scale numbers are the company's own: over 3 million developers, tens of thousands of enterprise customers, and more than 40 million generated models. Those are usage claims, not audited figures, but the partner list is concrete enough to show the model is past the research stage.
The money behind the model
VAST's growth has been backed by capital. The company's B and B+ rounds, announced on September 1, 2026, totaled about ¥3 billion (≈ US$420M / HK$3.3B) according to Xinhua Finance and the Beijing E-Town authority, with investors spanning venture, industry, and state-linked funds; VAST said its fundraising across the first half of 2026 reached roughly ¥5 billion in total. The funding race reflects a wider bet: that 3D generation, not just image and video, becomes a default creative layer for games, film, XR, and embodied-AI training.
That last point is why the category attracts both startups and giants. A 3D model that knows physics and contacts can feed robot simulators as readily as game engines, so the same asset pipeline serves two very different buyers.
Honest limitations
This article draws on VAST's own product announcements, developer conference presentations, and investment-trackers such as 睿兽分析 relayed by National Business Daily; the model-behavior claims and the usage counts are vendor-reported and not independently audited. The "20 billion parameters" and "under ten seconds" figures come from VAST's launch material and secondary tech press, not a third-party benchmark. The partner and integration names are stated by the company and were not confirmed with each partner here. The about-¥3 billion figure for the combined B and B+ rounds comes from Xinhua Finance, the Beijing E-Town authority, and securities press; it is a company-announced amount reported by journalists, not an audited regulatory filing, and the "roughly ¥5 billion half-year total" is VAST's own framing. We did not test Tripo 3.0 ourselves, and we have not assessed the licensing of its training data, the true rate of post-generation cleanup required in production, or how it compares head-to-head with rivals such as Rodin from Deemos (影眸科技). This covers one company's product, not the whole AI-3D field.
What readers can do now
- If you make games or 3D content, trial Tripo or a rival on your ugliest asset class — hard-surface props with symmetry — and measure cleanup time, not just first-impression quality.
- If you buy 3D-generation tooling, ask for the post-processing and rigging rates separately, and check that exports land in your engine's expected format before committing.
- If you follow the space, watch whether these models move from "single asset" to "whole scene" generation — that is the line between a faster prop shop and a new production pipeline.
