Documentation Navigation
Models & pricing / Models / MiniMax-H3-experimental
MiniMax-H3-experimental Overview
This model's data has not been fully verified and benchmarks may be missing.
MiniMax-H3-experimental is a chat-reasoning AI model created by Kijai, featuring undisclosed parameters and a context window of unknown.
An experimental model using 4-bit weights and int8-convrot activations, with a VAE that speeds up decode times by ~1.5x, built by Comfy-Org.
Model Lineage & Specification
minimax-h3-experimental is Kijai's primary release in the current family.
minimax-h3-experimentalComparable models
| Feature | MiniMax-H3-experimental | Solar Pro 4 | Qwen3.8 Max Preview | Nemotron 3 Ultra |
|---|---|---|---|---|
| Description | An experimental model using 4-bit weights and int8-convrot activations, with a V... | Solar Pro 4 is a large language model from Upstage built for enterprise workflow... | Preview Qwen flagship for million-token multimodal reasoning and long-horizon ag... | NVIDIA Nemotron 3 Ultra is a flagship open-weight frontier LLM family engineered... |
| API Identifier | minimax-h3-experimental | upstage-solar-pro-4 | alibaba-qwen3.8-max-preview | nvidia-nemotron-3-ultra |
| Parameters | — | — | — | 550B total / 55B active |
| Context Window | unknown | 524288 | 1M tokens | 1M tokens |
| License / Type | open weights | api only | api only | open weights |
MiniMax-H3-experimental
An experimental model using 4-bit weights and int8-convrot activations, with a V...
minimax-h3-experimentalSolar Pro 4
Solar Pro 4 is a large language model from Upstage built for enterprise workflow...
upstage-solar-pro-4Qwen3.8 Max Preview
Preview Qwen flagship for million-token multimodal reasoning and long-horizon ag...
alibaba-qwen3.8-max-previewNemotron 3 Ultra
NVIDIA Nemotron 3 Ultra is a flagship open-weight frontier LLM family engineered...
nvidia-nemotron-3-ultraAn experimental model using 4-bit weights and int8-convrot activations, with a VAE that speeds up decode times by ~1.5x, built by Comfy-Org.