xAI's Grok 4.6 release and benchmarks

xAI's Grok 4.6, released August 12, 2026, scores 61 on the Artificial Analysis Intelligence Index, up five points from Grok 4.5 and level with GPT-5.6 Sol Max, one point below Fable 5 Max. It is priced at $2/$6 per million input/output tokens, 80% and 88% cheaper than Fable 5 Max's $10/$50. On agentic benchmarks, it reaches 1,753 Elo on GDPVal-AA v2, above both GPT-5.6 Sol Max (1,728) and Fable 5 Max (1,741), though Artificial Analysis notes overlapping confidence intervals with Fable 5 and Qwen3.8 Max. On AA-Briefcase it reaches 1,577 Elo, in the Fable 5 tier but behind the Claude Opus 5 family. On APEX-Agents it scores 57.5%, narrowly above GPT-5.6 Sol Max (56.7%) but below Fable 5 Max (59.2%). Terminal-Bench v3.0 shows a larger gap: Grok 4.6 at 26% versus 34.6% for GPT-5.6 Sol Max and 34.1% for Fable 5 Max. On coding, it scores 69.9% on CursorBench v3.2, edging Sol Max's 67.2%, though it trails both Sol and Fable on DeepSWE and Terminal-Bench. Grok 4.6 is notably turn-efficient, averaging about 53 turns and 0.5 billion input tokens on AA-Briefcase versus about 103 turns and 2.0 billion for Claude Opus 5 Max. The model is available via API, Cursor, Grok Build, OpenRouter, Vercel, and Cloudflare. xAI attributes gains to a longer training run, regenerated SFT trajectories, model-based trace filtering, agentic RL across coding, web development, CAD, and kernel optimization, and more self-testing during long tasks. Elon Musk said Grok 4.7 is significantly better than 4.6, initial training is complete, and supplemental training on SpaceX company data is underway, with an expected timeline of 3-4 weeks from the launch announcement.