SpaceXAI Releases Grok 4.7: A Bigger Base Mannequin on the Similar $2/$6 Value as Grok 4.6

SpaceXAI has launched Grok 4.7, its new flagship mannequin for coding, agentic duties, and information work. Grok 4.7 is constructed on a bigger base mannequin and an extended reinforcement studying run. It nonetheless ships on the identical worth and velocity as Grok 4.6.

Is it deployable? Sure, as a hosted mannequin. You may name grok-4.7 in the present day by the xAI API, Cursor, Grok Construct, OpenRouter, Vercel, and Cloudflare.

What Modified Below the Hood

SpaceXAI lists 4 adjustments over Grok 4.6:

1. A brand new, bigger base mannequin: Grok 4.7 doesn’t reuse the Grok 4.6 base.
2. An extended RL run on more durable duties: The duty combine is weighted towards issues that take many hours to finish.
3. Higher self-verification and long-context dealing with: The corporate says the mannequin checks its personal work extra fastidiously.
4. Native Grok Bot harness assist: It was skilled to know the Grok Bot harness for conversational and information work.

The developer docs listing the API specs:

Benchmarks

The launch desk compares Grok 4.7 at xHigh effort with Grok 4.6 Excessive, GPT-5.6 Sol Max, and Fable 5.1 Max. The Grok 4.7 DeepSWE rating was run at excessive effort. All scores are vendor-reported.

*Excessive effort

Grok 4.7 improves on Grok 4.6 in each row. The most important leap is on Terminal-Bench 4.0, from 20.3% to 38.0%. EEBench rose 11 factors to 64.0%, the highest rating within the desk. On Harvey’s authorized agent benchmark, Grok 4.7 scored 19.6% towards 6.7% for Fable 5.1 Max.

Grok 4.7 doesn’t lead throughout the board. Fable 5.1 Max tops 4 of seven benchmarks, together with a 57.9% Terminal-Bench rating. GPT-5.6 Sol Max holds the highest DeepSWE v1.1 consequence at 72.7%.

Value is the opposite axis. Fable 5.1 Max prices 5x extra on enter and about 8.3x extra on output. GPT-5.6 Sol Max prices 2x extra on enter and about 3.3x extra on output. On a CursorBench 4.0 cost-per-task chart, SpaceXAI locations Grok 4.7 on the frontier in price-performance.

On GDPval, which assessments skilled information work, Grok 4.7 xhigh scored 1,695 Elo. That’s up from 1,605 for Grok 4.6 excessive. Fable 5.1 max leads at 1,735, and GPT-6 Astra max scored 1,542. SpaceXAI additionally says Grok 4.7 is best at creating paperwork and shows.

Security and Cybersecurity

Grok 4.7 ships with a completely new safeguard stack. SpaceXAI calls it the strongest mannequin it has examined on refusals and jailbreak resistance. It topped LatchBio’s biosafety benchmark at 62.4%.

On HackerBench v0.3, SpaceXAI’s personal benchmark for dangerous and malicious cyber duties, the mannequin let 3.3% of dangerous dual-use prompts by. The corporate says it not often blocks reputable safety work. Choose cybersecurity companions now get invite-only entry to its red-team capabilities for protection analysis.

Pricing and Availability

Grok 4.7 prices $2 per million enter tokens and $6 per million output tokens. It’s accessible in Cursor on all plans and is the default mannequin in Grok Build. It is usually served by the Grok API, OpenRouter, Vercel, and Cloudflare.

Grok 4.7 Quick is similar mannequin on quicker infrastructure, with twice the output velocity at twice the value. The docs say it runs solely in Cursor and Grok Construct, not on the general public xAI API. It is usually excluded from Grok Construct’s free tier.

A US regional endpoint at https://us.api.x.ai/v1 retains inference in the US at a ten% premium. SpaceXAI recommends setting a prompt_cache_key for dependable cache hits.

import os
from xai_sdk import Consumer
from xai_sdk.chat import consumer

shopper = Consumer(api_key=os.getenv("XAI_API_KEY"))
chat = shopper.chat.create(mannequin="grok-4.7")
chat.append(consumer("Explain this repo."))
print(chat.pattern().content material)

Interactive Explainer

“;toks.forEach(operate(t){h+=’

‘+t[0]+’

‘});
toks.forEach(operate(tq,q){h+=’

‘+tq[0]+’

‘;
toks.forEach(operate(tk,okay){var a=okay(q,okay);h+=”})});
$(“#grid”).innerHTML=h;
$$(“.rowl”).forEach(operate(r){var f=operate(){selq=+r.dataset.q;drawMask()};r.addEventListener(“click”,f);r.addEventListener(“keydown”,operate(e){if(e.key===”Enter”||e.key===” “){e.preventDefault();f()}})});
var t=toks[selq],seen=toks.filter(operate(_,okay){return okay(selq,okay)}).map(operate(x){return x[0]}),seesX=seen.some(operate(s){return s[0]===”X”});
var msg;
if(mode===”full”){msg=”

Everything sees everything

“+t[0]+” attends to all 9 tokens, including the noisy latent X. Because X changes at every step, the prefix keys and values change too, so nothing can be cached.

“}
else if(t[2]===”x”){msg=”

“+t[0]+” is the image being denoised

It attends to “+seen.be part of(“, “)+”. The latent reads the full prefix and every patch of its own block, so it gets complete context from text and references.

“}
else{msg=”

“+t[0]+” is part of the prefix

It attends to “+seen.be part of(“, “)+”. It “+(seesX?”sees”:”never sees”)+” the noisy latent X, so its keys and values stay identical across all denoising steps. That is what makes the prefix KV cache valid.

“}
$(“#mexp”).innerHTML=msg;dimension()}
$$(“#seg button”).forEach(operate(b){b.addEventListener(“click”,operate(){$$(“#seg button”).forEach(operate(x){x.classList.take away(“on”)});b.classList.add(“on”);mode=b.dataset.m;drawMask()})});
drawMask();

/* ———- panel 4: decision + rgba ———- */
var ars=[[“1:1”,2048,2048],[“4:3”,2400,1792],[“3:4”,1792,2400],[“3:2”,2528,1696],[“2:3”,1696,2528],[“16:9”,2752,1536],[“9:16”,1536,2752]];
$(“#ar”).innerHTML=ars.map(operate(a,i){return ‘‘+a[0]+’‘}).be part of(“”);
operate setAr(i){var a=ars[i],w=a[1],h=a[2],m=210,s=m/Math.max(w,h);var f=$(“#frame”);f.fashion.width=Math.spherical(w*s)+”px”;f.fashion.top=Math.spherical(h*s)+”px”;
$(“#px”).textContent=w+” x “+h;$(“#mp”).textContent=(w*h/1e6).toFixed(2)+” MP”;$(“#lat”).textContent=(w/16)+” x “+(h/16)}
$$(“#ar .chip”).forEach(operate(c){c.addEventListener(“click”,operate(){$$(“#ar .chip”).forEach(operate(x){x.classList.take away(“on”)});c.classList.add(“on”);setAr(+c.dataset.i)})});
setAr(0);
var star=$(“#frame svg path”),dot=$(“#frame svg circle”);
$$(“#bg button”).forEach(operate(b){b.addEventListener(“click”,operate(){$$(“#bg button”).forEach(operate(x){x.classList.take away(“on”)});b.classList.add(“on”);var f=$(“#frame”),v=b.dataset.b;
f.classList.toggle(“checker”,v===”checker”);f.fashion.background=(v===”checker”)?””:(v===”alpha”?”#000″:v);
if(v===”alpha”){star.setAttribute(“fill”,”#fff”);star.setAttribute(“stroke”,”#fff”);dot.setAttribute(“fill”,”#fff”)}else{star.setAttribute(“fill”,”url(#g)”);star.setAttribute(“stroke”,”#2B2A7A”);dot.setAttribute(“fill”,”#fff”)}})});
$(“#copy”).addEventListener(“click”,operate(){var t=”This is an RGBA image with transparency. . The image has alpha channel and the background is transparent.”;
var performed=operate(){$(“#copy”).textContent=”Copied”;setTimeout(operate(){$(“#copy”).textContent=”Copy”},1400)};
if(navigator.clipboard&&navigator.clipboard.writeText){navigator.clipboard.writeText(t).then(performed,performed)}else{performed()}});

/* ———- panel 5: ship ———- */
var V=[
[“ok”,”Allowed”,”The license grants a royalty-free, worldwide right to use, copy, modify, and distribute the model for non-commercial purposes, which it defines as research or evaluation.”],
[“stop”,”Needs a separate commercial license”,”Commercial use is not covered. Qwen asks teams to request a license at model-business@notice.qwencloud.com before shipping.”],
[“warn”,”Allowed for non-commercial use, with conditions”,”If you use the model or its outputs to build and release another AI model, show u201cBuilt with Qwenu201d or u201cImproved using Qwenu201d in its docs. You cannot use u201cQwenu201d as the primary product name.”],
[“warn”,”Allowed, with notices”,”Give recipients a copy of the license, mark files you changed, and keep the Qwen attribution notice in a Notice file. The non-commercial limit still applies.”]];
operate verdict(i){var v=V[i];$(“#verdict”).innerHTML=”;dimension()}
$$(“.use”).forEach(operate(u){u.addEventListener(“click”,operate(){$$(“.use”).forEach(operate(x){x.classList.take away(“on”)});u.classList.add(“on”);verdict(+u.dataset.u)})});
verdict(0);
dimension();
})();