The AI landscape just got a little more crowded at the top. On September 21, 2026, xAI released Grok 4.7 its most capable model yet for coding, agentic workflows, and professional knowledge work. And while it's not quite dethroning the current frontier leaders, it's a serious statement of intent from Elon Musk's AI company.
So, what's actually new?
Grok 4.7 is built on a much larger foundation 2.1 trillion parameters, a 40% jump from Grok 4.6's 1.5 trillion. It also comes with a 500,000-token context window, meaning it can hold entire codebases or lengthy documents in memory during a single session. One quirky but telling detail: xAI trained the model on supplemental data from SpaceX, including satellite telemetry and engineering failure logs, aiming for a model that reasons better about real-world hardware and physical systems.
Where it really shines
The biggest gains show up in coding and long-task performance. On Terminal-Bench 4.0, Grok 4.7 nearly doubled its predecessor's score, jumping from 20.3% to 38%. CursorBench 4.0 rose from 40.4% to 46.3%, and DeepSWE v1.1 climbed from 65.2% to 71%. xAI also trained the model to natively work with Grok Bot, its persistent AI agent for businesses, a sign that xAI wants Grok 4.7 to function less like a chatbot and more like a system you can assign a task to and walk away from.
The catch
Despite the upgrades, independent benchmarks tell a more nuanced story. Grok 4.7 improves on every metric compared to Grok 4.6, but it doesn't top the charts overall. Claude's Fable 5.1 still leads on several coding and long-task evaluations, and GPT-6 Astra edges it out on electrical-engineering benchmarks. In short: Grok 4.7 is stronger, but it's playing catch-up rather than setting the pace.
Pricing and access
Despite the bigger model, pricing hasn't budged: $2 per million input tokens and $6 per million output tokens, same as Grok 4.6. There's no wait-list it's live now in Cursor, Grok Build, and the xAI API, with more integrations expected in the coming weeks.
Why it matters
For developers, this is a model that's explicitly built to handle multi-hour, complex tasks with less hand-holding a real shift from single-reply chatbot interactions toward genuine task delegation. For businesses, flat pricing and wide availability make it easy to test without a budget overhaul.
The bigger picture: the gap between AI labs is narrowing, and everyone xAI, Anthropic, OpenAI is racing to build models that don't just answer questions but actually get work done. Grok 4.7 is a strong step in that direction. Whether it's the step depends on which benchmark you're looking at.
