Grok 4.7 is now available on Vercel's AI Gateway at 40% off through September 27, according to Vercel's changelog dated September 21, 2026. SpaceXAI launched the model the same day, and Vercel's model page lists the promotional rate at $1.20 per million input tokens and $3.60 per million output tokens, against SpaceXAI's $2 and $6 list price.
The catch: nearly every performance number in the launch is SpaceXAI's own. Artificial Analysis, an independent tracker, posted a first score within hours, but its speed and pricing fields were still blank when I checked.
The launch also ended a wait. Elon Musk's early September countdown pointed to September 11, and on that date he wrote that the model "needs a few more days to cook," per CellCog's log of his posts. It shipped ten days after that target.
What Did SpaceXAI and Vercel Actually Announce?
Vercel confirms the model ID spacexai/grok-4.7, a 500K token context window, and four reasoning levels: low, medium, high and xhigh. The discount applies automatically, and Vercel says the gateway adds no markup or platform fee on inference, including on bring-your-own-key requests.
SpaceXAI's launch post describes a new, larger base model trained with a longer reinforcement learning run weighted toward multi-hour tasks. It starts at $2 per million input tokens and $6 per million output tokens, the same as Grok 4.6, and a fast variant offers twice the output speed at twice the price. The company's pitch is "Twice as fast, at half the price of comparable models."
Where Is Grok 4.7 Available, and What Does the Gateway Discount Cover?
SpaceXAI says Grok 4.7 is live in Cursor and Grok Build, and through its Grok API, third-party coding harnesses, model routers and cloud platforms. On Vercel, the same ID works with the AI SDK, the OpenAI-compatible Chat Completions API, and agents including fx, Cursor, Codex, Amp and OpenCode. Zero Data Retention and prompt-training opt-out are supported.
Vercel has not said who funds the discount or whether the fast variant is offered on the gateway. SpaceXAI's price is also a starting rate. Third-party tracker mem0 reports that Grok 4.6 bills at higher rates once a prompt reaches 200K tokens. I found no Grok 4.7 long-context rate, so test before feeding it very large prompts.
How Do SpaceXAI's Benchmarks Hold Up?
SpaceXAI chose the comparison set and ran Grok 4.7 at xhigh effort against Grok 4.6 at high, so treat every score as vendor-reported.
Source: SpaceXAI launch post, September 21, 2026.
Grok 4.7 beats Grok 4.6 on all seven benchmarks in SpaceXAI's full table and leads this field on EEBench and the Harvey legal benchmark. It trails Fable 5.1 by 5.5 points on CursorBench 4.0 and by nearly 20 points on Terminal-Bench 4.0.
Two things do not line up. SpaceXAI lists GPT-5.6 Sol at $4 and $20 per million tokens, but Artificial Analysis's August 12 review of Grok 4.6 listed it at $5 and $30, and I could not reconcile the two. The twice-as-fast comparison is also against other vendors, while the same post says Grok 4.7 matches Grok 4.6's speed.
Artificial Analysis scores Grok 4.7 (xhigh) at 46 on its Intelligence Index v4.3.2 and says the run used 240M output tokens against a 92M median, which it calls very verbose. Musk had promised better token efficiency than Grok 4.6, though xhigh inflates token counts and I have not seen Artificial Analysis's Grok 4.6 figure. SpaceXAI has not published a parameter count, so Musk's 2.1 trillion figure stays unconfirmed.
Why It Matters
This is the third AI Gateway model addition I have covered in seven days, after Gemini 3.8 Live models on September 15 and GLM 5.3 FlashX on September 18. Gateways are turning launches into same-day, one-line swaps, and promotional pricing is the lever that gets you to try the new one. For anyone already routing through a gateway, the switching cost is a string change.
The risk is mistaking launch-week pricing for a price. Cost per finished task matters more than cost per token, and no independent cost-per-task figure for Grok 4.7 exists yet. Musk's timing record is one more reason to wait: CellCog logged four separate release timelines from him before launch, none on SpaceXAI's site.
What to Watch
- Artificial Analysis filling in Grok 4.7's speed, pricing and cost-per-task fields. That is where the twice-as-fast claim gets tested.
- What Vercel does on September 28. Its no-markup policy suggests the price returns to $2 and $6, but Vercel has not said, and an extension is possible.
- Whether SpaceXAI discloses a parameter count.
- Cloudflare's separate AI Gateway product. Its xAI provider docs show Grok 4 in examples and its other Grok pages name 4.3 and 4.20 variants. I found no Grok 4.7 entry there today.
Related BriefFlash coverage
Model routing is where enterprise AI is getting hard, as our look at the enterprise AI operations problem shows.
By Alex Carter. Published September 21, 2026. The news broke September 21, 2026.
Corrections: any change to the facts above will be logged here in the format "Updated [date]: [what changed and why]."
Key Takeaways
- Grok 4.7 is available on Vercel's AI Gateway at 40% off through September 27, priced at $1.20 per million input tokens and $3.60 per million output tokens against SpaceXAI's $2 and $6 list price.
- Vercel confirms a 500K token context window and four reasoning levels (low, medium, high, xhigh). SpaceXAI confirms availability in Cursor, Grok Build and its own Grok API.
- Independent verification is thin: Artificial Analysis scores the model 46 on its Intelligence Index v4.3.2 but has not filled in speed or pricing, and SpaceXAI's benchmark table is vendor-reported.
FAQ
Where is Grok available?
Grok itself is available on grok.com, the iOS and Android apps, and inside X, according to the navigation on SpaceXAI's site. For Grok 4.7 specifically, SpaceXAI's launch post says it is available today in Cursor and Grok Build, through the Grok API, and through third-party coding harnesses, model routers and cloud platforms. Vercel added it to AI Gateway on September 21, 2026. The launch post does not say whether Grok 4.7 has reached the consumer apps, so do not assume it has.
Does Grok have its own browser?
I found no browser made by SpaceXAI. Xplorer and Grok Browser are independent, Chromium-based open source projects built by outside developers around Grok, not SpaceXAI products. Grok Build did gain a Browser Use plugin that lets it drive a real browser, according to Browser Use's own post on X, but that is an agent tool rather than a standalone browser.
How do I get access to Grok?
For Grok 4.7, several routes are confirmed. You can create an API key in the SpaceXAI console, try it free in Grok Build using the install command SpaceXAI links from its launch post, select it in Cursor, or call it through Vercel AI Gateway with the model ID spacexai/grok-4.7. On Vercel's playground, users who have not made a payment get $5 of credits every 30 days. SpaceXAI's launch post does not say which consumer Grok plans include Grok 4.7.
How do I enable Grok 4.7 on AI Gateway?
In code, create an AI Gateway API key and pass spacexai/grok-4.7 as the model ID in the AI SDK or through the OpenAI-compatible Chat Completions API. For coding agents, install the latest Vercel CLI with npm i -g vercel@latest, run vercel ai-gateway setup, then select the model in fx, Cursor, Codex, Amp or OpenCode. To start a new eve agent with xhigh reasoning, run npx eve@latest init grok-agent --model spacexai/grok-4.7 --reasoning xhigh. The 40% discount applies automatically through September 27.
