Meshy MCP 0.6.10 switches Smart Topology's default to T2.5, changing 3D game asset budgets
By GameSlash
The November 9 update changes the default model and polygon count used by AI game-asset agents, and adds validation for topology and image ratios.
Meshy MCP Server released version 0.6.10 on November 9, 2026, changing the default Smart Topology model for Text to 3D and Image to 3D from Meshy T2 to T2.5 when an agent does not explicitly specify ai_model. This matters to game developers because existing prompts may now generate a higher-polygon asset without changing their instructions. Official release notes

Why game makers should check polygon budgets after updating
Previously, unspecified Smart Topology requests used meshy-t2, whose API default was 4,000 faces. Version 0.6.10 now selects meshy-t2.5, with an API default of 15,000 faces. If you are making mobile games or scenes filled with props, specify target_polycount instead of assuming your earlier triangle budget will be preserved.
For example, set model_type to "smart-topology", ai_model to "meshy-t2.5", and target_polycount to 5000 for roughly a 5,000-face target. Specify ai_model = "meshy-t2" if you need the previous model. T2.5 supports 100–30,000 faces and generates triangle meshes rather than quads. Detailed changelog

Two other fixes for asset workflows
Image to 3D now rejects a Smart Topology request for quad topology instead of silently ignoring the setting. If you need editable quad loops in Blender, use meshy_remesh afterward. For A-pose or T-pose generation, the image tool also explains when certain Nano Banana models return a 3:4 image despite a different requested aspect ratio.
Upgrade path and limitations
You can install or update with npx @meshy-ai/meshy-mcp-server@latest and restart your MCP client, following the project's instructions. Meshy API model creation requires an API key and billed credits under the applicable plan. This is a change to the MCP server's defaults, not the first launch of the T2.5 model. GameSlash has not independently benchmarked the outputs.
Sources: Meshy MCP Server 0.6.10 release · Official character model comparison