PRODUCT UPDATE / 3D assets · Smart Topology
Meshy T2.5 joins the API with Smart Topology targets up to 30,000 faces
T2.5 is now the default Smart Topology model for image and text generation. Its default target is 15,000 faces, with unchanged credit pricing and triangle-only output.
For developers generating game props in batches and maintaining Meshy API integrations.
On October 15, 2026, Meshy added T2.5 to its Image to 3D and Text to 3D APIs and made it the default model for Smart Topology generation. It accepts targets of 100–30,000 faces at the same API credit prices as T2. Developers generating game props in batches should check their model selection and polygon budget before the next run.
The default target rises from 4,000 to 15,000 faces
The Image to 3D documentation now specifies T2.5 when a request sets model_type: "smart-topology" but omits ai_model. The two models have different budgets:
| Model | Target faces | Default |
|---|---|---|
| Meshy T2 | 100–15,000 | 4,000 |
| Meshy T2.5 | 100–30,000 | 15,000 |
To retain the previous model, explicitly set ai_model: "meshy-t2". When moving to T2.5, setting target_polycount also lets background props and close-up assets keep their own budgets. The requested count is approximate; actual geometry can land above or below it.
Meshy describes T2.5 as preserving sharper detail for game assets. Both Smart Topology models produce triangles, and requesting quad topology returns an error. T2.5 accepts a single image or text; Multi-Image to 3D does not support it.
Mesh generation uses the face budget directly
Smart Topology generates directly toward the target count and skips a subsequent Remesh pass. The documentation says should_remesh and decimation_mode are ignored in this mode. The Text to 3D API uses the same ranges and defaults.
Meshy's low-poly prop cookbook illustrates the workflow with an oak tree image. A 1,000-face target with texturing disabled produced the pictured gray mesh of 1,096 triangles. Those measurements date to September 21, 2026, before T2.5. The example explains the face target and texture setting; it does not establish the new model's quality.
Mesh-only generation still costs 5 credits
The current API pricing table lists T2.5 Image to 3D at 5 credits for geometry alone, 15 credits total with 2K or 4K textures, and 20 credits total with 8K textures. Text to 3D charges 5 credits for its Preview mesh stage, with the Refine texture stage billed separately. Requests use prepaid API credits.
Older integrations also need to check for model_type: "lowpoly". Meshy's documentation schedules its retirement for October 30, 2026, after which those requests will fail, and recommends migrating to Smart Topology with T2.5. Updating a batch script is a useful point to make its model, face target and texture setting explicit.
Sources (6)
No sponsorship or affiliate links in this article.
