Hunyuan3D
3D Available Full comparison ↗ tencent/hunyuan3d · by Tencent
· diffusion-3d-dit
Pricing — 6 offering(s)
About this offering: Tencent's Hunyuan3D 3.1, resold by Atlas Cloud at flat per-generation prices in USD. Rapid costs $0.30 for base geometry ($0.50 with PBR materials), and Pro (labelled v3.1) costs $0.50 for a textured model, rising to $0.90 with every add-on. The same model is also sold by fal, WaveSpeed AI and Tencent Cloud.
Rapid text/image-to-3D (v3.1)
- $0.30 / generation
Current
2026-09-04 → present
Hunyuan3D 3.1 Rapid on Atlas Cloud: $0.30 per generation, or $0.50 with PBR materials, for text-to-3D or…
Hunyuan3D 3.1 Rapid on Atlas Cloud: $0.30 per generation, or $0.50 with PBR materials, for text-to-3D or image-to-3D. Rapid is the faster edition, taking about 2–3 minutes per model with 1K textures (Atlas Cloud pricing page, checked 2026-10-11).
Pro text/image-to-3D (v3.1)
- $0.50 / generation
Current
2026-09-04 → present
Hunyuan3D 3.1 Pro on Atlas Cloud in textured (Normal) mode: $0.50 per generation, $0.70 with PBR materials or…
Hunyuan3D 3.1 Pro on Atlas Cloud in textured (Normal) mode: $0.50 per generation, $0.70 with PBR materials or a custom face count, and $0.90 with both. An untextured geometry-only model costs $0.30, or $0.50 with a custom face count. Pro is the full model: about 3–6 minutes, up to 4K PBR textures and 40K–1.5M polygons (Atlas Cloud pricing page, checked 2026-10-11).
About this offering: Hunyuan3D 3.1 on fal, priced separately from fal's v2.x endpoint. Two editions: Rapid ($0.225 per generation; PBR materials add $0.15) for speed, and Pro ($0.375; PBR, multi-view input and custom face count each add $0.15) for quality. Against fal's v2.0 tiers, Rapid costs about 1.4x the untextured $0.16, and Pro costs less than the textured $0.48. Open weights on Hugging Face.
Rapid text/image-to-3D (v3.1)
- $0.225 / generation
Current
2026-09-02 → present
Hunyuan3D 3.1 Rapid on fal: $0.225 per generation for text-to-3D or image-to-3D, plus $0.15 for PBR…
Hunyuan3D 3.1 Rapid on fal: $0.225 per generation for text-to-3D or image-to-3D, plus $0.15 for PBR materials. Rapid is the speed-optimised edition (fal model page, checked 2026-10-11).
Pro image-to-3D (v3.1)
- $0.375 / generation
Current
2026-09-02 → present
Hunyuan3D 3.1 Pro on fal: $0.375 per generation, with PBR materials, multi-view input and a custom face count…
Hunyuan3D 3.1 Pro on fal: $0.375 per generation, with PBR materials, multi-view input and a custom face count each adding $0.15, so a run with all three costs $0.825. Pro is the quality-optimised edition (fal model page, checked 2026-10-11).
About this offering: Tencent's own Hunyuan3D API (Tencent HY 3D Global), 3.1 generation, billed in credits at $0.02 each postpaid ($0.0135–$0.015 prepaid). The Express edition uses 15 credits per call ($0.30 postpaid) and the Professional edition 25 credits ($0.50), with PBR, multi-view input and custom face count adding 10 credits each. Tencent names editions rather than a version number, but Professional matches the 3.1 Pro model sold by other hosts.
Express (Rapid) text/image-to-3D (v3.1)
- $0.02 / credit
Current
2026-09-04 → present
Tencent HY 3D Global (Express) API — 15 credits/call for text or image input ($0.30 effective at the…
Tencent HY 3D Global (Express) API — 15 credits/call for text or image input ($0.30 effective at the $0.02/credit postpaid rate; prepaid packs bring the credit rate to $0.0135–$0.015). +10 credits ($0.20) to enable PBR materials. Express is the latency-optimised edition (~2–3 min, 1K textures, fixed mid-range poly budget). Failed tasks are not charged.
Professional (Pro) text/image-to-3D (v3.1)
- $0.02 / credit
Current
2026-09-04 → present
Tencent HY 3D Global (Professional) API, "Normal" GenerateType — 25 credits/call for a textured geometry…
Tencent HY 3D Global (Professional) API, "Normal" GenerateType — 25 credits/call for a textured geometry model ($0.50 effective at $0.02/credit postpaid). Other GenerateTypes: Geometry (white model, no texture) 15cr ($0.30), LowPoly 30cr ($0.60), Sketch 25cr ($0.50). Each of MultiViewImages / EnablePBR / FaceCount adds 10cr ($0.20). Post-processing calls are billed separately: UV Unfold 10cr, Auto Rigging 10cr, Smart Topology 50cr, 3D Part Generation 30cr, 3D Texture Edit 30cr, Convert 3D Format 5cr. Pro is the full model (~3–6 min, up to 4K PBR, 40K–1.5M polygons). Failed tasks are not charged. Billing doc last updated 2026-05-08.
About this offering: Tencent's Hunyuan3D 3.1, Rapid edition, resold by WaveSpeed AI at a flat $0.27 per generation with no texture or complexity surcharge. That's a little more than fal's Rapid ($0.225) and a little less than Atlas Cloud's ($0.30). The 3.1 line is served through hosted APIs; the open weights cover the earlier 2.x versions.
Rapid text/image-to-3D (v3.1)
- $0.0225 / generation
Historical
2026-09-04 → 2026-10-10
WaveSpeed's earlier flat rate for Hunyuan3D 3.1 Rapid, for text-to-3D or image-to-3D, with no surcharge for…
WaveSpeed's earlier flat rate for Hunyuan3D 3.1 Rapid, for text-to-3D or image-to-3D, with no surcharge for mesh complexity or texture quality.
- $0.27 / generation
Current
2026-10-11 → present
A flat $0.27 per generation for Hunyuan3D 3.1 Rapid, for text-to-3D or image-to-3D, with no surcharge for…
A flat $0.27 per generation for Hunyuan3D 3.1 Rapid, for text-to-3D or image-to-3D, with no surcharge for mesh complexity or texture quality (WaveSpeed model pages, checked 2026-10-11). WaveSpeed raised it from $0.0225 at some point after 2026-09-04.
About this offering: Hunyuan3D 2.x on fal, priced as two versions: v2.0 at $0.16 untextured or $0.48 textured, with fine-grained parameter control, and v2.1 at $0.30, which fal now marks deprecated. fal's multi-view and turbo v2 variants aren't listed here. fal prices the newer 3.1 line separately.
Image-to-3D, white mesh (v2)
- $0.16 / generation
Current
2026-08-31 → present
Hunyuan3D 2.0 on fal: $0.16 per generation for an untextured (white) mesh, with adjustable octree resolution…
Hunyuan3D 2.0 on fal: $0.16 per generation for an untextured (white) mesh, with adjustable octree resolution, inference steps and guidance scale (fal model page, checked 2026-10-11).
Image-to-3D, textured mesh (v2)
- $0.48 / generation
Current
2026-08-31 → present
Hunyuan3D 2.0 on fal: $0.48 per generation for a textured mesh, three times the untextured price (fal model…
Hunyuan3D 2.0 on fal: $0.48 per generation for a textured mesh, three times the untextured price (fal model page, checked 2026-10-11).
Image-to-3D (v2.1 architecture)
- $0.30 / generation
Current
2026-08-31 → present
Correction
Corrected 2026-10-11: Modelglass first recorded a wrong price for these dates; this row replaces it, and the…
Corrected 2026-10-11: Modelglass first recorded a wrong price for these dates; this row replaces it, and the wrong row is no longer served.
Hunyuan3D 2.1 on fal: $0.30 per generation. fal marks this endpoint deprecated and says the model is no longer supported, so use Hunyuan3D 3.1 for new work (fal model page, checked 2026-10-11).
About this offering: Tencent's open-weights 3D generator, hosted by several providers. These are Replicate's prices for v2.0: about $0.10 per run at default settings, billed per second of GPU time, so it varies with the inputs. Versions so far are 2.0 (January 2025), 2.1, 3.0 and 3.1 (Pro and Rapid editions); 3.1 is sold by fal, Atlas Cloud, WaveSpeed AI and Tencent Cloud.
Image/Text-to-3D (Hunyuan3D 2.0)
- $0.024 / generation
Historical
2026-08-31 → 2026-10-10
Replicate's earlier per-run estimate for v2.0 at default settings, about $0.024 (41 runs per $1). Replicate…
Replicate's earlier per-run estimate for v2.0 at default settings, about $0.024 (41 runs per $1). Replicate bills per second of GPU time, so the cost per run varies with the inputs.
- $0.10 / generation
Current
2026-10-11 → present
Replicate's current estimate is about $0.10 per run (10 runs per $1) at default settings. It's billed per…
Replicate's current estimate is about $0.10 per run (10 runs per $1) at default settings. It's billed per second of GPU time on an Nvidia L40S at $0.000975 per second, and a run typically takes up to about 106 seconds, so the cost varies with the inputs (Replicate model page, checked 2026-10-11).
Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.
See how Hunyuan3D fits into a cost-aware routing setup
See how →Capability profile
How Hunyuan3D rates across core capability dimensions, with the task-level evidence behind each rating.
Solid production-grade meshes from a single image; 3.x notably improved. Below Rodin's fidelity ceiling but well ahead of earlier open models. No independent geometry benchmark; not currently placed on the 3D Arena top ranks.
Image-to-3D primary and reliable; text-to-3D supported. Multi-view input (v2.1+, up to 8 views in 3.1) tightens fidelity to a reference.
Dedicated texture-synthesis model (Hunyuan3D-Paint). Textured output is a 3x price multiplier on most hosts. PBR support varies by host endpoint.
Dense triangle mesh output; octree resolution / inference steps are configurable on some endpoints but there's no quad / low-poly mode.
No built-in rigging or animation.
Text, image, and (v2.1+) multi-view. Varies by host endpoint.
Host-dependent: Replicate v2.0 ~25s on an L40S; the 3.x "Rapid" edition ~2-3 min; "Pro" slower. Turbo variants exist on fal.
Operator guidance
The value pick among 3D generators, and the only major one you can self-host. Hosted prices start at $0.10 per run on Replicate (v2.0) and $0.16 untextured or $0.48 textured on fal (v2.0); the 3.1 line costs $0.225 (Rapid) or $0.375 (Pro) on fal and a flat $0.27 for Rapid on WaveSpeed AI. Choose Hunyuan3D when cost or open weights matter and you don't need built-in rigging, segmentation or fine topology control. Step up to Tripo for the fuller pipeline, or Rodin for fidelity.
Use cases
- Low-cost per-generation 3D from a single image at scale, from $0.10 per run on Replicate
- Self-hosted / on-prem 3D generation (open weights)
- Pipelines that only need geometry + basic texture, no rig/segmentation
Limitations
- No built-in rigging, animation, part segmentation, or quad topology
- Per-generation price and capabilities vary substantially by host and version
- No independent quality benchmark; not on 3D Arena's top ranks
- v2.x (fal + Replicate) and 3.1 (fal) are priced; 3.0 and non-fal 3.x hosts (WaveSpeed, Tencent Cloud) are not yet tracked