Grok 4.7 lands at $2/$6 — a cheap flagship that trails on coding
xAI's new flagship ships today on the API, Cursor and Grok Build with a 500K context, but Terminal-Bench puts it behind even DeepSeek V4.1 Flash.
Copy markdown
A $2/$6 flagship you can call today
Grok 4.7 is live on the xAI API at $2 per million input and $6 per million output, with a 500K-token context and a May 2026 knowledge cutoff. xAI built it on a larger base, trained it with longer RL, and tuned it to verify its own output — its pitch is code plus general agentic work.
The coding gap is real — price it accordingly
On Terminal-Bench 4.0 agentic coding it scores 26%, behind GPT-6 Astra's 60% and edged out by DeepSeek V4.1 Flash at 27%; on the Artificial Analysis Intelligence Index it lands at 46 versus 53 for Claude and GPT-6. Reach for it as a cheap generalist, not your coding lead.
Day one in Cursor and Grok Build
You don't have to touch the raw API — Grok 4.7 is selectable in Cursor and Grok Build from launch, so you can A/B it against your current model in the editor before you commit. At $2/$6 it matches Qwen3.8 Max on price.
Elsewhere: Heretic strips refusals in one command
pip install heretic-llm then `heretic <model>` runs optimized abliteration on most dense and several MoE models. Its Gemma 3 cut suppressed refusals as hard as hand-tuned versions (3 of 100) at far less capability damage (0.16 KL divergence vs up to 1.04), in ~20–30 minutes on a 3090. AGPL-3.0.