xAI Launches Grok 4.6: Enhanced Long-Running Agents, Matches GPT-5.6 on AA Index
FinanceYF5 · x · 2026-08-13
xAI has officially released Grok 4.6, focusing on enhancing long-running agents to handle complex tasks like modifying entire codebases and turning ideas into polished applications. It scores 61 on the Artificial Analysis Intelligence Index, matching GPT-5.6 Sol, and surpasses competitors on the GDPVal-AA knowledge work benchmark.
Training & Capabilities:
- Data & Training: Underwent a longer supplemental training run than v4.5, incorporating reasoning and advanced engineering data. SFT trajectories were regenerated using Grok 4.5, with RL covering general coding, kernel optimization, web dev, and CAD.
- Coding Benchmarks: DeepSWE increased from 54% to 65.9%, CursorBench from 66.7% to 69.9%, and FrontierCode from 56.6% to 61.3%.
- Availability & Pricing: Now available on Cursor, Grok Build, API, and more. Standard pricing remains unchanged ($2/M input, $6/M output), while the Fast version is double the price. Cursor and Grok Build offer 2x included usage for the first week.
Related event: xAI Releases Grok 4.6 with Major Performance Leap at Same Price(37 posts)→
More from coding & agent
- Developer Builds Pac-Man Clone Using Only Grok for Code, Music, and Art — bennash · 2026-08-13
- Geek Demo: Driving Raspberry Pi Controlled Phones with Gemini and CLI Tools — hugs · 2026-08-13
- 13 Free Official Claude AI Courses: From Basic Prompting to Advanced MCP — Aiden_Tech_Ai · 2026-08-13
- SF AI Infrastructure Night to Explore Open Local Agent Stacks and SaaS Data Security — _changxu · 2026-08-13
- AI Coding Agents Cause Analytic Abundance: Verification Becomes the New Bottleneck — arjunrajlab · 2026-08-13
- Open-Source Three.js Game Agent Skills Hits 1.3k Stars — majidmanzarpour · 2026-08-13