3D generation
Text- and image-to-3D models ranked by blind comparison arenas, the apps that generate and texture ready-to-use assets, and node-based pipelines that run open 3D models on your GPU.
Models are ranked by Rapidata Image-to-3D Leaderboard (benchmark.ai, Overall), 3D Arena (Hugging Face), Render view, then Top 3D AI Arena: Textured (monthly season) for models without a result on the first. Tools inherit the tier of the best model they offer. How we rank.
- Best overallSHi3D v3.01,366 on Rapidata Image-to-3D Leaderboard (benchmark.ai, Overall)
- Best valueAMeshy 7.1Tier A for $0.6/generation
- Best free startSHi3DFree tier · paid plans · via Hi3D v3.0
- Best local (≤24 GB GPU)CTRELLIS.2 (4B)~24 GB VRAM · 1,005 on Rapidata Image-to-3D Leaderboard (benchmark.ai, Overall)
- Tier: how close to the best result on the task's main public benchmark
- Outlined: a tool inherits the tier of the best model it offers
- No public result yet
- Cloud · Local: runs on your own hardware
Models
The engines themselves. Use them through an API or inside the apps on the right.
Apps
Ready-made apps with a UI: sign in or install, then start.
Workflow platforms
Node and canvas tools that chain several models into one pipeline.
Runtimes & APIs
Run open models on your own machine, or reach models and services through an API.
Nothing tracked yet
Quality vs. setup
The same tools as the map, placed by how good the results can be (up) and how much setup they take (right). Top-left is the easiest way to top quality; further right buys control, privacy or price at the cost of setup. Models are left out: you reach them through these tools.
- Cloud
- Local: runs on your hardware
- Hollow: tier inherited from the best model the tool offers
Not placed (no public result yet): Alpha3D, Autodesk Flow Studio, Lychee Gen, Neural4D, Sloyd, Spline, Triverse AI, VARCO 3D, YVO3D
Everything, in order
| Tier | Name | Approach | Runs | Rated by | Price | Skill |
|---|---|---|---|---|---|---|
| S | Hitem3D (Math Magic) | Models | Cloud | 1,366 on Rapidata Image-to-3D Leaderboard (benchmark.ai, Overall) | — | — |
| S | Hitem3D (Math Magic) | Apps | Cloud | via Hi3D v3.0 | Free tier | beginner |
| S | Scenario | Apps + Workflow platforms | Cloud | via Hi3D v3.0 | Free tier · $15/mo | intermediate |
| A | Meshy | Models | Cloud | 1,299 on Rapidata Image-to-3D Leaderboard (benchmark.ai, Overall) | $0.6/generation | — |
| A | 3D AI Studio | Apps + Workflow platforms | Cloud | via Meshy 7.1 | Free tier · $19/mo | beginner |
| A | Meshy | Apps | Cloud | via Meshy 7.1 | Free tier · $20/mo | beginner |
| B | Tripo v3.1 VAST | Models | Cloud | 1,261 on Rapidata Image-to-3D Leaderboard (benchmark.ai, Overall) | — | — |
| B | Krea | Apps + Workflow platforms | Cloud | via Tripo v3.1 | Free tier · $5.25/mo | beginner |
| B | VAST | Apps | Cloud | via Tripo v3.1 | — | beginner |
| C | Tencent | Models | Cloud | 1,145 on Rapidata Image-to-3D Leaderboard (benchmark.ai, Overall) | $0.5/generation | — |
| C | ByteDance Seed | Models | Cloud | 1,098 on Rapidata Image-to-3D Leaderboard (benchmark.ai, Overall) | — | — |
| C | Microsoft | Models | Local | 1,005 on Rapidata Image-to-3D Leaderboard (benchmark.ai, Overall) | ~24 GB VRAM | — |
| C | Tencent | Models | Local | 751 on Rapidata Image-to-3D Leaderboard (benchmark.ai, Overall) | ~29 GB VRAM | — |
| C | Tencent ARC Lab | Models | Local | 755 on Rapidata Image-to-3D Leaderboard (benchmark.ai, Overall) | Open weights | — |
| C | Deemos (Hyper3D) | Models | Cloud | 957 on Rapidata Image-to-3D Leaderboard (benchmark.ai, Overall) | — | — |
| C | Meta | Models | Local | 648 on Rapidata Image-to-3D Leaderboard (benchmark.ai, Overall) | ~32 GB VRAM | — |
| C | Tencent | Apps | Cloud | via Hunyuan3D 3.1 | — | beginner |
| C | Comfy Org | Workflow platforms | Local | via TRELLIS.2 (4B) Cloud models: tier A | Free | advanced |
| C | Nikhil Shahane (Bingeljell) | Apps | Local | via TRELLIS.2 (4B) | Free | intermediate |
| C | Piotr Wilkin | Runtimes & APIs + Apps | Local | via TRELLIS.2 (4B) | Free | advanced |
| C | Igor Aherne | Apps | Local | via TRELLIS.2 (4B) | Free | intermediate |
| C | Comfy Org | Workflow platforms | Cloud | via TRELLIS.2 (4B) | Free tier · $20/mo | advanced |
| C | Zimeng Xiong | Apps | Local | via Hunyuan3D 2.1 | Free | beginner |
| C | Deemos | Apps | Cloud | via Rodin Gen-2.5 | $30/mo | beginner |
| – | Alpha3D | Apps | Cloud | No public result yet | — | beginner |
| – | Alpha3D | Models | Cloud | No public result yet | — | — |
| – | Autodesk | Apps | Cloud | No public result yet | Free tier | beginner |
| – | Lychee Gen Mango3D | Apps | Cloud | No public result yet | — | beginner |
| – | Lychee Gen 2 Mango3D | Models | Cloud | No public result yet | — | — |
| – | DreamTech | Apps | Cloud | No public result yet | Free tier · $20/mo | beginner |
| – | DreamTech | Models | Cloud | No public result yet | — | — |
| – | Sloyd | Apps | Cloud | No public result yet | Free tier · $15/mo | beginner |
| – | Sloyd | Models | Cloud | No public result yet | — | — |
| – | Stability AI | Models | Local | No public result yet | ~7 GB VRAM | — |
| – | Spline | Apps | Cloud | No public result yet | $15/mo | beginner |
| – | StepFun | Models | Local | No public result yet | ~27 GB VRAM | — |
| – | Triverse AI | Models | Cloud | No public result yet | — | — |
| – | Triverse AI | Apps | Cloud | No public result yet | Free tier · $20/mo | beginner |
| – | NC AI | Apps + Workflow platforms | Cloud | No public result yet | — | beginner |
| – | NC AI | Models | Cloud | No public result yet | — | — |
| – | Autodesk | Models | Cloud | No public result yet | — | — |
| – | YVO3D | Apps | Cloud | No public result yet | — | beginner |
| – | YVO3D | Models | Cloud | No public result yet | — | — |
Benchmarks used
- Rapidata Image-to-3D Leaderboard (benchmark.ai, Overall) by Rapidata (blind human votes, Bradley-Terry score on the Elo scale (R = 400·β + 1000) from pairwise human preference on turntable renders of image-to-3D outputs) · snapshot 6 Oct 202613 models; 52,667 matches decided by 263,258 votes (5 per match, gold-standard-filtered annotators) on 357 fixed input images; ±17-20 95% CIs. Overall view used (Geometry and Appearance views exist). Header says updated 2026-10-06 (a footer says 2026-10-18). Image-to-3D only; no Tripo P-series, SPAR3D or Step1X-3D.
- 3D Arena (Hugging Face), Render view by 3D Arena (Dylan Ebert, Hugging Face) (blind human votes, Elo from pairwise community votes on rendered image-to-3D outputs (Render view)) · snapshot 18 Oct 202624 entries from the Space's /api/leaderboard?vote_type=render JSON (a Topology view ranks 21). Image-to-3D only; closed entries are unversioned. Has TRELLIS.2, Hunyuan3D 2.1, SAM 3D, Pixal3D, SPAR3D and Meshy 7 (not 7.1); no Tripo, Rodin, Hunyuan3D 3.x, Seed3D or Hi3D. No date shown.
- Top 3D AI Arena: Textured (monthly season) by Top 3D AI (blind human votes, Online Elo from blind community votes (Textured board); ratings reset to 1000 every monthly season) · snapshot 19 Oct 2026Oct 2026 season, in progress (about 330-1,080 votes per entry so far). The page now labels this board 'Overall Quality' (Overall ELO); a Geometry Fidelity rating is also published per entry. Entries are product generators at a stated mode, incl. Alpha3D 5, Lychee Gen 2, YVO3D 2.5, VARCO 3D 2.0, Sloyd AI v3, Triverse 3.0, N4D-2.5, Autodesk Wonder3D. Separate Low Poly, Segmentation, Smart UV boards.