Tripo P1
Last reviewed
May 16, 2026
Sources
20 citations
Review status
Source-backed
Revision
v1 ยท 3,013 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 16, 2026
Sources
20 citations
Review status
Source-backed
Revision
v1 ยท 3,013 words
Add missing citations, update stale details, or suggest a clearer explanation.
Tripo P1, marketed in full as Tripo Smart Mesh P1.0, is a production grade native 3D diffusion model developed by the Beijing based company Tripo AI, the consumer brand of VAST. The model was previewed at the 2026 Game Developers Conference in San Francisco in mid March 2026 and announced publicly on 25 March 2026, alongside a 50 million US dollar funding round backed by Alibaba and Baidu Ventures. Tripo P1 is positioned as a real time graphics oriented sibling to Tripo H3.1, the company's high fidelity model, and is built on a unified three dimensional probabilistic space rather than the autoregressive token streams used by earlier generative 3D systems.
The model is trained on a corpus of roughly 50 million high quality 3D assets and is reported by Tripo AI to generate production ready polygon meshes in as little as two seconds, an improvement the company describes as up to 100 times faster than its earlier mesh generation pipelines. It targets game engines, robotics simulation, and extended reality, where engine ready topology and predictable polygon counts matter more than maximum visual detail. The headline rival systems are Tencent Hunyuan 3D, Deemos Rodin Gen-2, and Meshy 6, and Tripo P1 sits alongside these as one of the most discussed releases of the 2026 commercial 3D generation cycle.
VAST was founded in March 2023 in Beijing, with a Cayman Islands holding structure, and publicly launched the Tripo product line in 2024. The company was co founded by chief executive Simon Song Yachen, a former co founder of the Chinese large language model company MiniMax and a former member of the chief executive office at SenseTime, together with Guoli Su. Song has said in interviews with the South China Morning Post and VoxelMatters that Google's 2022 release of the DreamFusion text to 3D paper was the trigger for starting VAST, since it was the first time he felt that 3D content generation had reached a point where it could realistically be made available to mainstream creators.
By late 2025 VAST employed more than 100 people across offices in Beijing and Hangzhou, reported annual revenue of roughly 12 million US dollars at an average of one million per month, and counted Tencent, ByteDance, HTC, and Stability AI among its enterprise customers. The team had also published more than 50 papers at CVPR, SIGGRAPH, and ICCV and open sourced 20 plus repositories including Wonder3D, ThreeStudio, and TripoSR, which together accumulated about 18,000 stars on GitHub. The company operates the Tripo Studio web platform at tripo3d.ai and a separately metered developer API at platform.tripo3d.ai.
The Tripo product line moved through several major releases before P1. TripoSR, an early image to 3D model produced jointly with Stability AI, was open sourced in March 2024 and provided much of the team's early visibility in the open source 3D community. The first hosted production model, Tripo v1, launched on the Tripo Studio web platform during 2024 and was followed by incremental upgrades through 2024 and 2025. Tripo 3.0, released in 2025, introduced the company's Sparse Flex representation, Standard and Ultra output modes, the Magic Brush 2.0 editing tool, and a Smart Part Segmentation feature that broke generated assets into separable components.
In the same product cycle the company published the Tripo H series, a high fidelity flagship model line that culminated in Tripo H3.1. The H series targeted detailed geometry, accurate texturing, and the kind of asset density needed for industrial design, high resolution 3D printing, and cinematic asset development. Tripo P1, introduced together with the refreshed H3.1 in March 2026, is the first member of a parallel production oriented family aimed at real time pipelines, and the company has also disclosed an early stage Tripo W1.0 world model initiative that uses related research for dynamic spatial environments.
The table below summarises the capabilities of Tripo P1 as disclosed in the Tripo AI press releases for the GDC 2026 debut, the 25 March 2026 funding announcement, and the company developer documentation. Capabilities not confirmed by a primary source are omitted.
| Capability | Status in P1 | Source |
|---|---|---|
| Native 3D diffusion architecture | Headline feature, unified probabilistic space | Tripo AI press release |
| Generation time per asset | Approximately two seconds for a production ready mesh | Tripo AI press release |
| Reported speed up over earlier Tripo pipelines | Up to 100x | Tripo AI press release |
| Training corpus size | Approximately 50 million high quality 3D assets | Tripo AI press release |
| Topology aware polygon mesh output | Supported, designed to skip retopology | Tripo AI press release |
| Low poly mesh generation | Headline use case, real time graphics ready | Tripo AI press release |
| Engine compatibility | Unity and Unreal Engine confirmed by company materials | Digital Journal coverage |
| Application targets | Game engines, robotics simulation, XR | Tripo AI press release |
| Image and text prompts | Supported through Tripo Studio | Tripo Studio documentation |
| Tripo Game Hub integration | Generated assets can be turned into interactive projects | Digital Journal coverage |
| Companion high fidelity model | Tripo H3.1 for industrial and cinematic detail | Tripo AI press release |
| Companion world model research | Tripo W1.0 early research initiative | The AI Insider |
Tripo AI describes the design intent for P1 as bridging the gap between generation and direct usability. The model is meant to produce assets that require minimal manual cleanup before being dropped into a game engine, which the company contrasts with prior generative 3D pipelines that emitted dense high polygon meshes needing a separate retopology step. Independent observers at Digital Journal note that the practical value of this claim will depend on how consistently P1 holds up across varied production scenarios rather than the controlled GDC demonstration scenes.
The disclosed architecture for Tripo P1 reflects a shift away from the autoregressive token by token mesh generation that dominated earlier generative 3D work. According to Tripo AI materials and statements by chief executive Simon Song, the P series models geometry directly within a unified three dimensional probabilistic space, with vertices, edges, and polygon faces represented within a shared spatial feature field. The diffusion process refines this entire field at once, allowing global topology and symmetry to emerge coherently rather than being assembled piece by piece.
Song framed the underlying motivation this way in the 25 March press release: "Three dimensional space is inherently holistic and symmetric. When geometry is forced into a sequence, artificial structure is introduced. Our approach models shapes directly in native spatial space, allowing structure to emerge coherently." In the same statement he positioned P1 as the realisation of that idea for real time production: "P1.0 is built around that idea, not just assisting existing pipelines, but becoming part of them."
The P series is reported to be trained directly on native polygon mesh data rather than on multi view image renderings, which Tripo AI argues is the reason that P1 can emit topology aware meshes that respect typical game engine expectations such as quad biased faces, reasonable polygon counts, and clean UV layouts. The company has not disclosed the parameter count of Tripo P1, the exact composition of the 50 million asset training corpus, or peer reviewed benchmark numbers, so much of the architectural discussion in independent coverage is still based on company statements rather than third party measurement.
Tripo AI also revealed two adjacent research tracks alongside P1. Tripo H3.1, the companion flagship, focuses on input alignment, geometry accuracy, and texture quality for complex assets such as characters and mechanical structures, and is the line aimed at industrial design, 3D printing, and cinematic work. Tripo W1.0 is an early stage world model effort that extends related ideas into dynamic spatial environments and is described by the company as preparatory research rather than a shipping product.
Tripo AI announced a 50 million US dollar funding round on 25 March 2026 in a press release distributed through PR Newswire and picked up by major industry outlets including 3D Printing Industry, The AI Insider, Pulse 2.0, Auganix, and Yahoo Finance. The round was backed by Alibaba and Baidu Ventures, and the company said the capital would be used to continue research on large scale 3D foundation models and to expand its global developer platform.
The announcement positioned the round as a strategic alignment with two of China's largest cloud and AI platforms. Alibaba operates the Aliyun cloud and the Tongyi family of foundation models, while Baidu runs the Wenxin large language model and has historically invested in spatial computing through Baidu Ventures. Neither investor disclosed a specific equity stake or post money valuation. Coverage in 3D Printing Industry and Pulse 2.0 framed the round as one of the larger pure play generative 3D investments of the year, and put it in the context of the broader 2026 competition between Chinese 3D model vendors including Tripo AI, Deemos, and Tencent's Hunyuan 3D team.
No prior priced rounds for VAST were disclosed in the March 2026 announcement, although VoxelMatters reported in 2025 that the company had achieved roughly 12 million US dollars in annual revenue and was operating profitably. Tracxn and other third party databases list earlier seed and angel activity under the VAST parent entity, but the company has not confirmed specifics through its own channels.
Tripo P1 is delivered through the Tripo Studio consumer web platform and through the separately metered Tripo developer API. Tripo Studio uses a tiered subscription model billed in US dollars and grants commercial usage of generated assets on all paid tiers; the free Basic plan does not permit commercial use. The table below summarises the published Tripo Studio plans as listed on tripo3d.ai/pricing, with the regular monthly price followed by the discounted monthly rate when billed annually.
| Plan | Regular monthly price | Discounted monthly (annual) | Monthly credits | Notes |
|---|---|---|---|---|
| Basic | 0 USD | 0 USD | 300 | One concurrent task, 20 stored models, no commercial use |
| Professional | 19.90 USD | 11.94 USD | 3,000 | Ten concurrent tasks, Smart Low Poly, three free retries per model |
| Advanced | 49.90 USD | 29.94 USD | 8,000 | Fifteen concurrent tasks, ten free retries, one free pro refine, 30 day history |
| Premium | 139.90 USD | 83.94 USD | 25,000 | Twenty concurrent tasks, unlimited retries, three free pro refines, unlimited downloads |
The annual billing tier carries a 40 percent discount versus monthly billing. All paid plans bundle access to Tripo v3.0 Ultra generation and to the company's auxiliary image generation tools with discounted credit rates. The Tripo developer API is a separate infrastructure layer and is sold under its own usage based contract rather than as an add on to the Studio subscriptions, which the company says is intended to keep large scale enterprise traffic separated from the consumer plans. Specific per credit API rates are negotiated directly with Tripo AI sales and are not published on the Studio pricing page.
Tripo AI did not publish a P1 specific price at the GDC 2026 debut. The model is included in the standard Tripo Studio credit ledger and the developer API, so the effective price of a P1 generation is set by the credit cost of the chosen output mode rather than by a separate P1 fee.
The table below places Tripo P1 next to three commonly cited rivals in the commercial generative 3D market as of May 2026. Figures come from each vendor's public documentation and from independent reviews when cited; values that have not been confirmed by a primary source are left blank.
| System | Vendor | Open weights | Headline architecture | Reported generation time | Headline strength |
|---|---|---|---|---|---|
| Tripo P1 | Tripo AI (VAST) | No | Native 3D diffusion, unified probabilistic space | About two seconds per asset | Engine ready low poly output, generation speed |
| Hunyuan 3D 3.0 | Tencent | Yes, community licence | Multi stage 3D generation with open checkpoints | Not directly comparable | Hard surface geometry, free open weights |
| Rodin Gen-2 | Deemos | No | 10 billion parameter native 3D foundation model, BANG part decomposition | One to two minutes per asset | Realism and part based generation |
| Meshy 6 | Meshy | No | Hosted generative 3D platform | Not directly comparable | Stylised characters, mature web product |
Reviewers at 3D AI Studio, Sloyd, and the Vset3D channel converge on a similar competitive picture. Tencent Hunyuan 3D is favoured for hard surface modelling and for the price of entry, since the weights are released under a community licence and can be self hosted within the licensed territory. Deemos Rodin Gen-2 is positioned at the realism and structural quality end of the market, with a 10 billion parameter native 3D foundation model and a recursive part based generation feature derived from the team's BANG paper. Meshy 6 is rated highly for product polish and security certifications including SOC 2 and ISO 27001.
Tripo P1 is positioned by the company and by early independent coverage as the speed and engine readiness leader in this group. Its appeal to game studios is the two second generation time and the engine compatible topology, which are useful when a team needs to iterate on dozens of background props per day. The trade off is that Tripo P1 is closed weight, runs only through Tripo AI infrastructure or the developer API, and has not yet published peer reviewed benchmarks that allow a direct measurement against Rodin Gen-2 or open weight Hunyuan checkpoints. Reviewers also note that the P series prioritises low polygon meshes, so workflows that need the highest possible visual fidelity may still be better served by the H3.1 companion model or by competitor systems with denser output.
Industry coverage of Tripo P1 has been broadly positive. The 3D Printing Industry write up of the 25 March funding announcement led with the 50 million dollar raise, the Alibaba and Baidu Ventures lead, and the two second generation time, and framed Tripo AI as one of the most active Chinese players in the 2026 generative 3D wave. The AI Insider focused on the technical pivot away from autoregressive token streams and quoted Song on the holistic spatial space argument. Pulse 2.0 emphasised the platform metrics: 6.5 million creators, 90,000 developers, and nearly 100 million 3D assets generated to date.
Games industry coverage centred on the GDC 2026 debut. Digital Journal described P1 as an attempt to close the gap between generation and engine integration by minimising the retopology and texture adaptation work that earlier generative 3D systems pushed onto artists, while noting that the effectiveness of the model would only be clear once it was used at scale in real game pipelines rather than the controlled GDC demonstration. TechTimes and Creative AI News framed P1 as a step into AI 3D 2.0, the marketing label that Tripo AI itself used in follow up press releases for the Smart Mesh P1.0 brand.
The most common reservations in independent coverage cover three areas. The first is that Tripo AI's headline numbers, including the 100x speed up and the 50 million asset training corpus, have not been independently benchmarked. The second is that the model is closed weight and only accessible through Tripo AI infrastructure, which contrasts with the open weight strategy that Tencent has used to grow the Hunyuan 3D ecosystem. The third is the gap between the GDC demonstrations, which leaned on stylised props and game assets, and the more demanding cases that show up in cinematic, industrial, or scanning grade workflows, where the companion Tripo H3.1 model is the recommended option.