Introducing Grok 4.6

What changed

Source summary

xAI released Grok 4.6, an upgrade to Grok 4.5 focused on long-running agentic tasks and more ambitious interactive and visual work, with improved performance on coding and domain-specific tasks such as kernel optimization and CAD, plus expanded safety calibration. It matches GPT-5.6 Sol on the Artificial Analysis Intelligence Index and is available via the API, Grok Build, Cursor, and partners including OpenRouter and Vercel, priced from $2 per million input tokens.

Why it matters to your work

Practical interpretation

If your team uses AI coding assistants or automation agents, Grok 4.6 offers stronger performance on complex, multi-step technical tasks — useful to test if you're evaluating agentic AI tools for engineering, coding, or ops workflows.

When this could help

Worth evaluating now, especially at $2/$6 per 1M tokens if you run agentic workloads where Grok 4.6's benchmark profile fits. First-week 2x usage promo in Grok Build and Cursor.

Grok 4.6 succeeds Grok 4.5 with improvements for sustained multi-step work. Benchmarks show it competitive with GPT-5.6 Sol and Fable 5: AA Intelligence Index 61 (Sol 61, Fable 62), GDPVal-AA 1753 (Sol 1728, Fable 1741), CursorBench 69.9% (Sol 67.2%, Fable 70.5%). Priced at $2/1M input, $6/1M output tokens.

How to get started

1. Access via the xAI API, Cursor, Grok Build at x.ai/build, or partners (OpenRouter, Vercel, Cloudflare). Select Grok 4.6 as your model.

   Provider guidance ↗

Build it into your routine

SCALA suggested workflow

1. Model comparison for agentic workloads

   When evaluating frontier models for a new agentic project, include Grok 4.6 in your benchmark comparison alongside GPT-5.6 Sol and Fable 5 — its cost-performance profile may suit sustained multi-step tasks.

What to check

• Trails GPT-5.6 Sol and Fable 5 on DeepSWE (65.9% vs 73%/70%) and Terminal-Bench (26% vs 34.6%/34.1%). Benchmark figures are mostly self-reported. Source ↗
• Subscription plan tiers for consumer access are not described. The fast variant costs double the standard rate. Source ↗