MiniMax

MiniMax H3 Max Spicy is a flagship uncensored video generation model powered by the high-tier MiniMax H3 Max architecture. Operating as a Spicy uncensored model, it completely bypasses standard safety guardrails to unlock maximum motion amplitude, intense physical interactions, and complex cinematic action sequences without content restrictions. Building on the enhanced rendering capacity and higher detail resolution of the Max tier, MiniMax H3 Max Spicy maintains exceptional frame consistency, realistic lighting, and photorealistic textures throughout dramatic camera movements. It is the premier tool for unrestricted dynamic video production, high-impact VFX pre-visualization, commercial action direction, and boundary-pushing creative AI visual assets.

MiniMax H3 Spicy is an advanced uncensored video generation model built upon the open-source MiniMax H3 architecture. Designed specifically as a Spicy uncensored model, it bypasses standard safety guardrails to synthesize fluid physical motions, dramatic camera pans, and intricate physical interactions without content restrictions. Preserving high frame consistency, crisp textures, and natural motion dynamics, MiniMax H3 Spicy efficiently powers unrestricted dynamic video creation, action sequence pre-visualization, and high-impact visual media pipelines.

MiniMax H3 MAX is an economical, fast video generation model from MiniMax (the team behind the Hailuo video engine). It supports text-to-video, first-frame-to-video, and first-and-last-frame-to-video generation with direct 768P output. Dual first/last frame control precisely locks the starting and ending visuals, with 4–15 second clips across aspect ratios from 21:9 to 9:16. Billing is fully transparent — from $0.05/second at 480P and $0.08/second at 768P, charged only for the output seconds actually generated, with input text and frame images completely free — making it a cost-effective choice for e-commerce showcases, social media videos, and bulk creative production.

MiniMax-M3 is MiniMax’s latest M-series multimodal foundation model, built for agentic reasoning, tool use, coding, and long-context tasks. It supports text, image, and video inputs with text output, offering a 1M-token context window, extended thinking, function calling, and structured outputs. With strong capabilities in long-horizon agent workflows, software development, multimodal understanding, and extended response generation, MiniMax-M3 is ideal for autonomous agents, coding assistants, document and video analysis, and production-grade applications that require massive context at a competitive cost.

MiniMax-M2.7 is a next-generation large language model built for autonomous real-world productivity and continuous improvement. Through advanced agentic capabilities and multi-agent collaboration, it can plan, execute, evaluate, and refine complex tasks in dynamic environments while actively contributing to its own evolution. Optimized for production-grade workflows, M2.7 excels at live debugging, root-cause analysis, financial modeling, and end-to-end document creation across Word, Excel, and PowerPoint. It achieves 56.2% on SWE-Pro, 57.0% on Terminal Bench 2, and a 1495 ELO rating on GDPval-AA, establishing a new benchmark for multi-agent systems operating in real-world digital workflows.