Grok 4.7 ships in Cursor at unchanged pricing — the AI coding contest has stopped being about price
xAI's Grok 4.7, released 21 September 2026 and available in Cursor on all plans from day one, jumped from 40.4% to 46.3% on CursorBench 4.0 while holding the same $2/$6-per-million-token pricing as its predecessor — a sign the AI coding tool race is now won on sustained task length and reliability, not sticker price.
24 September 2026
xAI released Grok 4.7 on 21 September 2026, calling it the company’s most capable model yet for coding and professional knowledge work, and it landed inside Cursor — on every plan, plus a faster variant — the same week. The pricing didn’t move: $2 per million input tokens and $6 per million output tokens, identical to Grok 4.6. What moved was capability. Grok 4.7 jumped from 40.4% to 46.3% on CursorBench 4.0, a benchmark built specifically around longer-running coding tasks, and from 65.2% to 71.0% on DeepSWE v1.1 at high reasoning effort. xAI says the gain came from a larger base model trained with a longer reinforcement-learning run, deliberately weighted toward problems that take many hours to finish rather than quick single-file completions.
Why this is a different contest than it looks like
For most of 2026, the AI coding tool story has been framed as a price war — cheaper tokens, aggressive discounting, vendors undercutting each other on cost per million tokens. Grok 4.7’s release doesn’t fit that frame: same price, meaningfully more capability, specifically on the metric that matters for commissioning work — how long a model can stay productive on a real task before it needs a human to intervene. Cursor now routes between Claude, GPT, Gemini and Grok models inside one interface, which means the tool a developer opens each morning is increasingly a thin shell over whichever underlying model currently leads on the tasks that matter to them, swapped in without changing how the tool is used.
Why this matters if you’re commissioning software
The practical upshot for a founder or CTO evaluating a development partner isn’t which brand of AI tool they use — it’s whether the partner is actually tracking which models are ahead on sustained, real-world tasks and using them, rather than being locked into whatever they set up eighteen months ago. Model-agnostic tooling is now table stakes for a team that wants to stay on the current capability frontier.
So what
If a development partner is still pitching a single AI tool by brand name as their whole differentiator, that’s a weaker signal than a partner who can explain why they’re using a specific model for a specific class of work this month. See our AI-assisted development approach, or get in touch to talk through how we scope a build around whichever tools are actually ahead right now.