SpaceXAI drops Grok 4.7: claims it beats GPT-5.6 Sol, but is it just more Musk-math?
Elon Musk and SpaceXAI finally pushed out Grok 4.7, a coding-focused model that supposedly outshines GPT-5.6 Sol. It’s definitely not a total disaster, though the release schedule seems to follow the classic "Musk time" dilation effect.
Grok 4.7 arrived with a flurry of internal benchmarks, hitting 46.3% on CursorBench 4.0, effectively nudging past its competitors. While it still trails slightly behind GPT-5.6 Sol on the DeepSWE v1.1 coding challenge, it managed to leap from 20.3% to 38% on the notoriously difficult Terminal-Bench 4.0.
Security measures also got a polish, with the model allegedly filtering out 96.7% of sketchy prompts while maintaining a fairly loose grip on legitimate cybersecurity research. Pricing remains locked at $2 per million input tokens and $6 for output, keeping it competitively positioned for developers who enjoy living on the bleeding edge.
Originally promised for late August, the model suffered through a mid-September delay, finally landing on September 21st after being sent back to the lab to "cook" a bit longer. Naturally, the hype cycle is already pivoting, with Elon Musk busy dangling the promise of Grok 4.8 and 4.9 like a carrot to keep the community from asking too many questions about today's version.
The tech industry continues its descent into a state of perpetual beta-testing where benchmarks are selected like artisanal cheeses to make every new release look like a revolution. It is truly inspiring how quickly a deadline can transform from a hard commitment into a mere suggestion once the marketing department needs to fill the void.
Source: x.ai
Comments
This is where the magic happens: AI reads your discussion and rewrites the article based on the most interesting comments. Each strong comment adds points to the meter below. Once the meter is full, the article updates live — no page reload needed.