SpaceXAI launched Grok 4.7 following a series of delays centred around multimodal support. Grok 4.7 brings with it a larger base model size than Grok 4.6, alongside considerable other improvements.
These improvements are primary focused on coding and knowledge work, with longer reinforcement learning on hard multi-hour tasks, tighter self-checking, and better use of long context windows. The updated model uses a 500k context window, a May 2026 knowledge cutoff (with search, as usual). The updated model also supports multimodal text and image input, with text output.
Grok 4.7 is now live via API, Cursor (all plans), and Grok Build, as well as through routers and gateways including OpenRouter, Vercel, and Cloudflare. For the moment, it is not available via Grok Bot as far as users can tell, which automatically picks the most suitable model for the task.
What Improved Over 4.6
One of the biggest improvements with Grok 4.7’s larger model size is the fact that its pricing has not been impacted. That means more difficult tasks, with better self-checking to prevent errors, can be had without any significant pricing increases.
Alongside the nearly doubled context window of 500K, Grok 4.7 will actually be cheaper to run over API or through raw token costs when factoring in the improvements through Grok Build’s ability to keep a warm context window on-server.
On company benches versus Grok 4.6, SpaceXAI reports CursorBench 4.0 at 46.3% versus 40.4%, DeepSWE at 71.0% versus 65.2%, EEBench at 64.0% versus 53.0%, AA Briefcase at 1,657 versus 1,546, Terminal-Bench at 38.0% versus 20.3%, Harvey Legal at 19.6% versus 15.8%, and HealthBench Professional at 56.7% versus 48.5%. Those are SpaceXAI’s numbers, not independent audits — but the pattern is a clear lift on coding agents, terminal work, and professional suites.
Same Price as 4.6 — Fast Is Different
Standard Grok 4.7 runs at the same pricing and speed as Grok 4.6. That means just $2 per million input tokens, and $6 per million output tokens. SpaceXAI positions it as twice as fast, at half the price of comparable models when looking at Opus 5.0 and Gemini 3.5, two older stalwarts.
Grok 4.7 Fast is a separate product, with roughly 2x the output speed, but at 2x the per token price: $4 per million input, and $12 per million output. It is currently only available in Cursor and Grok Build. SpaceXAI’s documents indicate that it should come to the xAI API within the coming days or weeks.
If you need to go extra fast on your work, you’ll need to pay for it. Interestingly, this is likely where some of Grok Bot’s tight constraints on usage limits come from. Grok Bot often defaults to using the Fast version of Grok, which results in higher usage in exchange for faster, more meaningful work.
Still Not Tip of Frontier
While the lift over 4.6 is still quite positive, especially for anyone regularly using Grok across Build, Bot, or in enterprise scenarios, it is still not quite a frontier model. The reported bench gains alongside unchanged API pricing means that Grok is still trailing three to six months behind current frontier models from Anthropic and OpenAI.
This release helps to close a gap between Grok 4.6 and the lighter offerings from frontier AI providers, and for now, doesn’t put Grok at the leading edge of the industry. In short, if you’re already using Grok across Cursor, Grok Build, or via API, it would be best to do your initial testing and begin moving over to Grok 4.7 for its increased intelligence and larger context windows.
