Latent Space AINews: xAI Launches Grok 4.6 Amidst Frontier Model Clashes
Latent Space · rss · 2026-08-13
This Latent Space AI newsletter covers major frontier model releases and ecosystem updates:
- xAI Grok 4.6: A 1.5T parameter model focusing on long-running agents and visual tasks. It achieves frontier-level price/performance with excellent Terminal-Bench results at a fraction of competitor costs. Elon noted Grok 4.7 training has begun.
- Open-Weight & Frontier Clashes: Alibaba open-sourced Qwen3.8-Max (2.4T MoE) focusing on long-context but lacking vision initially; DeepSeek V4 Pro GA enters the market at extremely low prices ($0.435/M input); Microsoft released its first from-scratch reasoning model, MAI-Thinking-1.
- Multimodal & Edge: LTX-2.5 open video stack upgraded with joint audio generation; Cohere released North Micro Vision (lightweight VLM); Google DeepMind introduced SL2T for sign-language-to-text.
- Inference & Compression: vLLM added support for Azure Blob and giant models; LLM Compressor introduced MoE expert pruning and arbitrary precision quantization, further lowering deployment barriers.
More from Infra
- New hardware day: local LLM deployment the moment it arrives — lxfater · 2026-10-16
- Together AI's score centering fixes RL training collapse under quantized inference — nrehiew_ · 2026-10-16
- Inference is splitting prefill and decode across specialized chips, not just H100s — AccBalanced · 2026-10-16
- Rebasely launches in beta: Git hosting built for humans and coding agents — handotdev · 2026-10-16
- Drama3 TTS lands in audio.cpp: up to 5x real-time on RTX 5090, <8GB VRAM mode too — Acceptable-Cycle4645 · 2026-10-16
- Three CoreWeave co-founders have sold $3.3B in shares since IPO amid credit worries — SumitGup · 2026-10-16