The Batch: Grok 4.6 Arrives, Claude Watermarks, Qwen3.8 Max Open Weights
DeepLearningAI · x · 2026-08-22
DeepLearningAI's latest The Batch newsletter covers: SpaceXAI's Grok 4.6 challenging top models, Anthropic adding invisible watermarks to Claude models, Alibaba's Qwen3.8 Max (2.4T params) open weights, and Agentic ASR for speech recognition. Also includes Andrew Ng's letter on AI engineering skills.
More from Models
- Frontier models got cheaper, a mystery 1M-context model went free: today in AI — Hesamation · 2026-08-22
- Grok 4.6's cache discount is only 75%, pricier than rivals for non-coding use — brandon_galang · 2026-08-22
- Test: Qwen 3.8-27B low preset outperforms older versions — Tall_Abrocoma_3533 · 2026-08-22
- AI agents wrote two 100K-word novels; readers found coherent plots but broken pacing — mrdrozdov · 2026-08-22
- Artificial Analysis benchmark praises Qwen 3.8 performance — Eyelbee · 2026-08-22
- OpenAI to cut GPT-5.6 Sol API pricing by over 20% for 3 months — Polymarket · 2026-08-22