MiniMax H3: local inference speed (tok/s), sourced
As of 2026-11-14, this dataset has 18 recipes for MiniMax H3. 10 were measured by the poster. These are image, video, or audio results. Speed is in the metric column, not tok/s. 6 entries have a date, from 2026-09-02 to 2026-10-10.
Filter these in the explorer →
| Model | Engine | Hardware | tok/s | Metric | Source level | Date | Source |
|---|---|---|---|---|---|---|---|
| MiniMax H3 (base, 50-step class) | ComfyUI | RTX 5090 | 1280×736×5s: 242 s → 168 s | Measured | link | ||
| MiniMax H3 vs FastH3 Turbo | Unspecified | RTX 5090 | 15 s clip: 11 min 55 s vs 2 min 6 s | Measured | link | ||
| MiniMax H3 (turbo) | ComfyUI | RTX 4090 | 5 s clip ~91 s; 15 s ~8 minutes; 960×544×5s about 50 s | Measured | link | ||
| MiniMax H3 | ComfyUI | RTX 3090 | 11 s video 13 minutes | Measured | link | ||
| MiniMax H3 (Sol accelerated / not accelerated) | NVIDIA Sol | 1× DGX Spark | Sol: 480p 5 s ≈ 68 s (NVIDIA published 56 s); without Sol >400 s; without Sol 720p 20 steps: 5 s 7 min / 10 s ~20 min / 15 s ~45 min | Measured | link | ||
| MiniMax H3 vs LTX-2.3 Eros | Unspecified | 1× DGX Spark | 864×480×5s: H3 average 270 s; LTX 154 s (1.76× faster) | Measured | link | ||
| FastH3 (4 steps) | MLX | MacBook Pro M5 Max 128GB | 1024×576×15 s: 43.7 → 13.3 minutes | Measured | link | ||
| MiniMax H3 | Unspecified | 48GB Apple Silicon Mac (chip not stated) | 5 s clip: 35 minutes → ~8 minutes | Measured | link | ||
| MiniMax H3 | TensorFold | RTX 5070 Ti 16GB + 64GB RAM (Windows) | 1344×768 5s with audio: 81 s (stock ComfyUI, 20 steps: 623 s) | Measured | 2026-10-03 | link | |
| MiniMax H3 | ComfyUI | Laptop RTX 5090 24GB + 64GB RAM | Three 5s chained clips: 17 min total | Measured | 2026-09-22 | link | |
| FastH3 V2 (MiniMax H3 8-step distillation) | Unspecified | RTX 5090 32GB | 832×480×5s: 14.8 s; 1344×768×5s: 38.6 s (with audio, end-to-end) | Official | 2026-10-06 | link | |
| FastH3 Trim (42 blocks pruned, experimental) | Unspecified | RTX 5090 | 480p 5s: 13.4 s; 768p: 35.4 s | Official | link | ||
| FastH3 V2 / Trim | Unspecified | RTX 4090 24GB | 480p 5s: 54.6 / 43.9 s; 768p: 154.6 / 132.8 s; 8GB limit Trim 82.0 s | Official | link | ||
| FastH3 V2 / Trim | Unspecified | 1× / 2× DGX Spark | 1×: 480p 141.4 / 125.8 s, Trim 768p 340.1 s; 2×: 87.2 / 78.3 s | Official | link | ||
| FastH3 Trim | MLX | Mac M4 Max 36GB | 480p 5s: 925.2 s (V2 not listed) | Official | link | ||
| FastH3 Preview (4 steps) | MLX | Mac M4 Max 36GB | 832×480×124 frames: cold start 465–504 s; INT6 456 s (peak 19.5GiB) | Official | 2026-09-02 | link | |
| MiniMax H3 Ref2VA (VSA + Turbo 4-step) | ComfyUI | RTX 4090 24GB | 1344×768×5.16s (24 fps / 124 frames): warm wall time ~72 s (first run ~95.2 s); peak VRAM ~13.5 GB. Same machine/seed/resolution: native dense 50-step ~650 s / ~22.5 GB; Turbo LoRA dense 4-step ~190 s / ~21.0 GB; VDN-H3 8-step ~135 s / ~18.5 GB (author table: first-run ~2.24× vs VDN; warm ~9× vs native) | Official | 2026-09-05 | link | |
| MiniMax H3 | LightX2V | RTX 5090 | 768×1344 15s video + audio ≈72 s (steady state, excl. MP4 export; baseline ~26 min) | Official | 2026-10-10 | link |
Numbers come from each source. This project did not measure them again. Models and settings differ between rows. Source levels: Official = project or vendor; Measured = the poster ran it (an engine or recipe author must also publish raw data and reproduction steps); Relayed = second-hand; Claimed = prediction or no reproduction details.