GMI 💚 @nvidia

Crazy. We gave GPT 6 Astra a video of our office, and in under an hour it built a nearly identical, fully explorable 3D version. Fable who?

5 more days of FREE MiniMax M3, M2.7, Music 3.0, and Speech 2.8 on GMI Cloud also, join the MiniMaxathon to get $500 cash, 3 months of MiniMax token plan max, and $200 GMI credits! Submit your work 👇

we are so backkkkkkkk huge thanks to @parker__conrad and @Support for helping us out in the past 48 hrs gonna bring more good news to our users in a few hours!

Getting hacked was bad. Spending an entire day trying to convince X @Support that our own company account actually belongs to us was worse. We haven’t slept all day. We’ve been trying to understand how @gmi_cloud was hacked and how to get our account back. We now believe it started with an extremely convincing phishing email impersonating X Support and asking us to “verify” ourselves. The email made it into our inbox without being flagged. We’ve raised this with our legal team, and they’re now handling the next steps. But what feels worst isn’t even just the hack. It’s how untrusted we’ve felt by X throughout the entire recovery process. We submitted two account recovery tickets. Both came back with: “We cannot verify you as the owner of the account.” Before we completely lost delegated access, we also tried reaching X Support directly. Same issue. No real help, because they couldn’t “verify” us. I’ve been managing @gmi_cloud all along. We can provide company documents, historical records, ad billing records — whatever is needed. As an AI company, we’ve always loved X. We build our community here, launch here, meet partners here. We genuinely believed this is the place to be. That’s why this feels less like anger and more like disappointment. X is too important of a platform for account recovery to work like this. When a real company loses access to its account, there needs to be a better way for the real people behind that company to be heard, verified, and helped 🙏

Took the sign to the streets of SF to talk to our customers

You can now use GMI endpoints in @vercel 5 models are available, including DeepSeek V4 Flash 0731 and Qwen 3.8 2.4 T, and Gemma 4, will add more later let us know what other models you wanna use GMI for on Vercel

Come moonlight with us! 🌙 Join @GoogleDeepMind, @ExaAILabs + some of our favourite AI companies for an exclusive AI hack lab in SF for top researchers, SWEs + technical founders. The challenge: launch your side hustle in one night. We will also have @DynamicWebPaige and more demoing! Build with @GoogleDeepMind, @ExaDevelopers, @WorkOS , @convex_dev, @gmi_cloud, @photonhq, @mosaic_so, @Vapi_AI, @elevenlabsio, @adalagent + @apify. Big prize pool, partner challenges + cash prizes on the night. RSVP 🧵

Introducing NVIDIA Nemotron 3.5 Lightning⚡ An open 30B MoE model with 3B active parameters, built for always-on agents to complete high-volume, specialized tasks faster. It delivers up to 4x the output speed of similar-sized models.

Seedance 2.5 is now live on GMI! 30% off for 48 hrs, making it almost the same price as MiniMax H3 one shot a 30-second scene with 50 references and production-ready editing

AgentBox version 2.0 is coming soon Its gonna be more agent-native: your agent would be able to able to start a container itself doing some internal tests now. what do you want our version 2.0 to include?

Proud to be the #1 provider for DeepSeek V4 Flash 0423 on @OpenRouter! Lowest input price ($0.038/M), ~40% cheaper than the #2 provider, and 24.6% of all daily tokens Thanks for building with us 🫡

DeepSeek V4 Flash 0731 is now available to all GMI users access via API key, onboarding to Open Router soon 👇

website update: now we had a dedicated section for developers including 1. Documentations 2. Demos 3. Overview Built something with GMI? Submit your project through the Demos page or link in 🧵 to be featured

💚 proud to be 1/230

Ran Claude Opus 5 in one marathon session for 24 hours straight. Ended up with this. Sakura Crossing: an explorable Japanese town in blossom season. And we open sourced it 👇