ICYMI: xAI releases Grok 4.6 for long-running agent work
xAI is pitching Grok 4.6 as an agent model for sustained project work, not just one-shot coding.
TestingCatalog says the release improves on Grok 4.5 for multi-step coding, research, visual, and interactive tasks. xAI attributes the gains to a longer training run, revised training methods, and reinforcement learning across coding and design environments. The model scored 61 on the Artificial Analysis Intelligence Index, with stronger results on some agent benchmarks and weaker results on DeepSWE and Terminal-Bench. Grok 4.6 is available through Cursor, Grok Build, the xAI API, OpenRouter, Vercel, and Cloudflare, with API pricing starting at $2 per million input tokens and $6 per million output tokens. TestingCatalog's note
TestingCatalog says the release improves on Grok 4.5 for multi-step coding, research, visual, and interactive tasks. xAI attributes the gains to a longer training run, revised training methods, and reinforcement learning across coding and design environments. The model scored 61 on the Artificial Analysis Intelligence Index, with stronger results on some agent benchmarks and weaker results on DeepSWE and Terminal-Bench. Grok 4.6 is available through Cursor, Grok Build, the xAI API, OpenRouter, Vercel, and Cloudflare, with API pricing starting at $2 per million input tokens and $6 per million output tokens. TestingCatalog's note
score 8