Introducing MiniMax M3: The First Open-Weights Model to Combine Three Frontier Capabilities - Coding & Agentic Frontier: 59.0% SWE-Bench Pro, 66.0% Terminal Bench 2.1, 34.8% SWE-fficiency, 28.8% KernelBench Hard, 74.2% MCP Atlas - MiniMax Sparse Attention scales context to 1M - Natively Multimodal from Step Zero API: Token Plan: 🚀New! MiniMax Code: Weights & Tech Report in ~10 Days

During the Minimax M3 free trial period, Ghast AI users across the platform have consumed over 2.6B tokens. From conversations to coding, users are putting Minimax M3 to work. with Ghast delivering a stable real-world experience. And we’re keeping the momentum going: starting today, Minimax M3 is now 50% off. Now is the best time to experience Minimax M3 in depth.

🔥 Exclusive: MiniMax M3 Limited-Time FREE Access is NOW LIVE! The wait is over! The reigning champion of the Artificial Analysis Open-Source Leaderboard, MiniMax M3, is now officially available on for a 7-day limited-time FREE trial! Don’t miss your chance to experience this high-performance powerhouse at ZERO cost. 🚀 🎁 Exclusive Bonuses Are Stacking Up: 1️⃣ New Users: Log in to via @BinanceWallet or @BitgetWallet to claim 1,000,000 FREE Credits instantly! 2️⃣ Deposit Match: Enjoy a 1:1 match for BNB Chain top-ups and 1:0.5 for other payment methods (Get up to $100 value in extra credits)! Stop hesitating—head over to and leverage top-tier open-source AI with @MiniMax_AI today! 👇 👉 Try it here:

MiniMax H3 × @fal 🎙️ This Thursday, join @OdinLovis, Creative Engineer at fal, Ethan Wei, AI Solutions Architect at MiniMax, and @VictorSuOrtiz, GTM Engineer at MiniMax for a live breakdown of how creators and developers are building with H3. We’ll cover multimodal references, native audio, prompting, LoRAs + customization, open weights, and real creative + technical workflows on fal. Aug 13 · 11 AM PT · X Spaces Set a reminder ↓

Minimax H3 🙂Realism People LoRA specialized in realistic people: faces that hold up in close-up, natural skin texture, believable expressions and gestures, film-style lighting and documentary camera movement. 👇

Minimax H3 is the first *open* video model I've seen that outperforms HunyuanVideo 1.5 - which is actually an impressive accomplishment for @TencentHunyuan to have held that throne for so long. For a while now I've been sad that the image/video frontier seemed to be getting much more closed than the text LLM frontier. H3 is a very strong step back in the open direction. This was generated in ~30 min on my AMD laptop:

MiniMax H3 ⚡️ Spectrum acceleration in ComfyUI — 34% lower Euler sampling time, 30% lower RES time Tested at approximately 0.5 MP, 8 seconds and 20 steps on an RTX PRO 6000 under WSL, using High VRAM mode 👇