Meshy raised nearly $400 million in a Series B on July 21, 2026, at a $1.5 billion valuation — the largest round any dedicated AI-3D company has closed. Tripo AI raised $200 million in a Series A/A+ on June 1, then $150 million more in a Series A3 on July 2. That's roughly $750 million into two text-to-3D companies in five weeks. None of it changes the thing that actually determines whether an asset lands in your engine clean: topology, UV layout, and collision geometry still need a human pass. The money is buying speed and scale, not a shortcut past retopology.
What $750 million actually bought in six weeks
Both raises came with product, not just a valuation bump. Meshy's pitch is control: Meshy 6 ships clean-topology options, a Low Poly Mode, and printability checks with one-click auto-repair before a mesh reaches a slicer or an engine importer. Tripo's pitch is throughput: its v3.1 pipeline generates a mesh in about 12 seconds against Meshy's under-two-minutes, plus auto-rigging and quad remesh baked into the same pass.
The clearest signal so far is a blind test published around Meshy's own comparison update on August 26: 1,331 senior 3D artists picked Meshy 6 over Tripo 3.1 in 63.8% of head-to-head comparisons on output quality. Tripo still wins on raw speed and volume — useful if you're generating fifty blockout props for a prototype, less useful if any of those fifty ship.
| | Meshy 6 | Tripo 3.1 |
|---|---|---|
| Generation time | Under 2 minutes | ~12 seconds |
| Blind-test preference (1,331 artists) | 63.8% | 36.2% |
| Notable feature | Printability checks, auto-repair, watertight export | Auto-rig, quad remesh |
| Best fit | Hero props, printable/game-ready candidates | Blockout volume, rapid iteration |
| July 2026 funding | $400M Series B ($1.5B valuation) | $350M combined (Series A/A+ and A3) |
Tripo's engine-facing push started earlier this year — Smart Mesh P1.0, shown at GDC on March 9, was built specifically to output native 3D diffusion meshes aimed at engine-ready use rather than sculptor-grade visual flex. That's the direction the funding is pushing both companies: fewer "impressive render" demos, more "drops into a project" claims.
Why funded doesn't mean game-ready
None of that changes what breaks when you actually import the output. Three things still fail the same way they did before either round closed:
Collision generation. An AI-generated mesh with inconsistent normals or non-manifold geometry produces the exact silent no-collision failure we covered when a convex hull generator chokes on inverted faces — the fix there (rebuilding a clean convex hull in the DCC before import) applies just as much to a Meshy or Tripo export as to a hand-modeled one.
Texture channel packing. Neither vendor's default export ties packed ORM channels to a UV layout your engine's material graph expects out of the box — the same Unity-vs-Godot-vs-UE5 mismatch we walked through on ORM textures shows up whether the source texture came from Substance or from a diffusion model.
Licensing provenance. Neither Meshy nor Tripo publishes a per-asset chain of custody for training data, so "AI-generated" and "clear for commercial shipping" are not the same claim — treat generated output the way you'd treat any unlicensed reference art until your legal review says otherwise.
Scale and pivot. Diffusion-based generation has no concept of your project's unit convention. A prop that looks correctly proportioned in the tool's own viewer routinely lands 40x too large or too small once it hits an engine with a different meters-per-unit default, and the pivot is almost never set to the object's base — both need a manual pass in Blender before the asset is worth importing anywhere.
How to actually use these tools in production today
The honest workflow right now: use Tripo for volume — blockout geometry, level-dressing filler, anything that never ships as a hero asset. Use Meshy for candidates you intend to clean up and ship, and budget real time for retopology and UV work regardless of which tool produced the base mesh. Before either goes into an engine, drop the GLB into BitSoul's 3D Studio to check tri count, UV overlap, and material slots against your budget — it runs in-browser on your own GPU, so you catch a broken UV seam before it's a bug ticket instead of after.
Treat the generator's output the same way you'd treat a rough sculpt from a contractor: a starting point, not a delivery. Run it through your normal pipeline gate — retopo pass, UV unwrap check, collision rebuild, scale verification against a reference prop — before it touches a scene file. If your team already has import presets for Blender, Unity, Godot, or UE5 through native plugins, point the cleaned mesh at those same presets rather than building a one-off import path just because the source was generative; the destination checks don't change based on where the mesh came from.
If you'd rather start from a mesh that's already collision-clean and channel-packed correctly, BitSoul's Combat Drone is a useful side-by-side reference for what "game-ready" looks like on a hard-surface prop of similar complexity to what these generators target. A free account's two monthly downloads cover evaluation; commercial use is included with paid memberships — see pricing.
The $750 million question isn't whether text-to-3D gets better — it will, and fast, given the capital now chasing it. It's whether "better" means fewer manual fixes per asset or just faster generation of assets that still need the same fixes. Right now, based on the blind-test data, it means the former for Meshy and the latter for Tripo. Pick per-asset, not per-vendor.
---
*This post is part of the Ultimate Guide to Free 3D Game Assets — BitSoul's complete reference for formats, texturing, rigging, optimization, and engine integration.*