SpaceXAI released Grok 4.6 on August 12, a post-training upgrade over Grok 4.5 that scores 61 on the Artificial Analysis Intelligence Index, tying GPT-5.6 Sol Max while maintaining pricing at $2/$6 per million tokens (MarkTechPost). The model ships 500K context and a new xhigh reasoning level. On agentic tasks, Grok 4.6 completes complex workflows in approximately 53 steps where Claude Opus 5 requires 103 steps, at a price more than 60 percent lower (The Decoder).
Dyna Robotics introduced Dyna-2, a world-action model pre-trained on more than one million hours of egocentric human video (MarkTechPost). The technical report establishes a scaling law on human data to 1M hours, the first transfer of that law to unseen robot data, and evidence that video co-training improves performance. Additionally, AllenAI published Open Instruct guidance for building custom LLM post-training pipelines using Supervised Fine-Tuning, Direct Preference Optimization, and Reinforcement Learning with Verifiable Rewards, optimized to run on 16GB hardware (MarkTechPost).