More posts
numbers Describe what the agent should say in plain language, drop in your product docs, done. Takes orders, books appointments, sends emails, hands off to a human when it gets stuck. ⚡️ Most voice AI is three separate pieces: speech-to-text, LLM, text-to-speech. Each step adds lag. Grok works on raw audio directly, one model, end to end. First word at 0.78 seconds. Full duplex, so it shuts up the moment you talk over it. 📂 Drag in PDFs, Word docs, or Markdown as your knowledge base. If you've got multiple agents, they all pull from the same FAQ. 💰 Custom voice or clone your own. $0.05/min. No platform fees, no surprises on the bill. 🚀 Starlink support is already using it. Call 888-GO-STARLINK and Grok answers. 🔗 x.ai/voice
supergrok via grok cli has a lower rate multiplier and comes with $200/mo in credits
. 1.6T params, ~48B active, native multimodal, aimed at long-horizon agents, trained start to finish on Chinese hardware. Images get quantized into multi-level discrete tokens, and that's where Visual Coding and 3D pick up. On the RL side, failed trajectories get turned back into verifiable training tasks. W4A4 inference, 43% faster prefill. Meituan's own evals put it between Claude Opus 4.8 and Opus 5.
It stacks layers instead of parameters. The decoder has 32 layers, but layers 4 through 27 run twice with shared weights, so a forward pass does 56 layers of work while the checkpoint stays at 9B. Their own eval numbers: 71.38 on BIRD-Dev and 13.72 on AutomationBench, both above Qwen3.5-27B. Terminal Bench is still behind.
