xAI releases Grok 4.6, focused on long-running agents, matching GPT-5.6 Sol on AA index
thione · x · 2026-08-18
xAI officially released Grok 4.6, building on Grok 4.5 with a focus on long-running agents and more ambitious interactive and visual work — staying with complex multi-step tasks like research, codebase-wide work, or turning an idea into a polished application.
Benchmarks: Matches GPT-5.6 Sol on the Artificial Analysis Intelligence Index (composite of nine benchmarks); frontier-level on GDPVal-AA, DeepSWE 1.1, CursorBench 3.2 and FrontierCode 1.1.
Training: A longer supplemental training run than 4.5 with curated model-generated reasoning data, high-quality engineering data, and an improved optimizer/recipe; SFT trajectories were regenerated with Grok 4.5 across reasoning efforts, agent harnesses and domains, filtered via model-based checks.
Available today in Cursor and Grok Build with 2x included usage for the first week.
More from coding & agent
- Paper refactors Agent skill protocol to solve context crowding — Zachly · 2026-08-18
- Warp launches coding agent for SSH with no extra CLI installs — vikvang1 · 2026-08-18
- Matt Pocock releases AI Coding Crash Course: Mastering Engineering Practices with Agents — mattpocockuk · 2026-08-18
- P2P Service over MCP Lets Users Query Each Other's Agents — plasticBarista · 2026-08-18
- Vibe Coding Lesson: AI Built It in 2 Weeks, One Dark-Mode Feature Took 4 Days to Refactor — dotey · 2026-08-18
- Google releases guide for migrating to latest Gemini models — rseroter · 2026-08-18