Grok 4.7 Launch: xAI's Coding Model Costs $2 per Million Tokens
Last updated: September 2026
Yesterday, September 21, 2026, SpaceXAI (xAI) released Grok 4.7, and the company calls it "our most capable model for coding and knowledge work." The model works longer on hard tasks, checks its own work more carefully, and ships with what xAI describes as its best-calibrated safety guardrails so far. The headline surprise is the price tag, because the new model arrives at exactly the same cost and speed as Grok 4.6.
Grok 4.7 is the new AI model released by SpaceXAI (xAI) on September 21, 2026. It targets coding and knowledge work, keeps the same price and speed as Grok 4.6, and costs $2 per million input tokens and $6 per million output tokens. You can try it today in Cursor, Grok Build, or the Grok API.
Source: xAI
The official benchmark table
xAI published a full comparison table alongside the launch, and it deserves a close read. Every number below comes straight from the official announcement page.
| Grok 4.7 xHigh | Grok 4.6 High | GPT-5.6 Sol Max | Fable 5.1 Max | |
|---|---|---|---|---|
| Input price $/M | $2 | $2 | $4 | $10 |
| Output price $/M | $6 | $6 | $20 | $50 |
| CursorBench 4.0 (software engineering) | 46.3% | 40.4% | 41.7% | 51.8% |
| DeepSWE v1.1 (software engineering) | 71.0%* | 65.2% | 72.7% | 70.0% |
| EEBench (electrical engineering) | 64.0% | 53.0% | 39.4% | 56.4% |
| AA Briefcase v1.1 (multi-hour office work) | 1,657 | 1,546 | 1,487 | 1,678 |
| Terminal-Bench 4.0 (terminal work) | 38.0% | 20.3% | 37.3% | 57.9% |
| Harvey Legal Agent (legal work) | 19.6% | 15.8% | 2.5% | 6.7% |
| HealthBench Professional (clinical work) | 56.7% | 48.5% | 60.5% | 62.1% |
(The asterisk marks the high-effort setting on DeepSWE.)
Three rows stand out immediately. On EEBench, which tests electrical engineering work, Grok 4.7 scores 64.0% while GPT-5.6 Sol Max manages only 39.4%. Terminal-Bench 4.0 nearly doubles from one generation to the next, rising from 20.3% on Grok 4.6 to 38.0% on Grok 4.7. And on Harvey Legal Agent, Grok 4.7's 19.6% sits far ahead of GPT-5.6 Sol Max at 2.5% and Fable 5.1 Max at 6.7%.
The overall GDPval Elo rating tells a similar story. Fable 5.1 (max) leads at 1735, Grok 4.7 (xhigh) sits close behind at 1695, Grok 4.6 (high) scores 1605, and GPT-6 Astra (max) reaches 1542. CursorBench 4.0 itself notes that Grok 4.7 stands at the frontier of price and performance for long coding tasks.
Same price as Grok 4.6, aggressive against rivals
The pricing is the real hook for most developers. Grok 4.7 keeps Grok 4.6's rates of $2 per million input tokens and $6 per million output tokens, and a fast variant doubles the output speed at double the price.
Compare that with the competition in the official chart. GPT-5.6 Sol Max charges $20 per million output tokens, roughly five times Grok's output cost. Fable 5.1 Max charges $50, which is more than eight times as much. xAI's own launch description sums it up as "twice as fast, at half the price of comparable models," and the company is careful to frame that comparison against similar competitors, not against Grok 4.6.
The rivals named in xAI's official chart give a sense of how crowded the field has become: Fable 5.1, Opus 5, GPT-5.6 Sol, Sonnet 5, GPT-6 Astra, Gemini, and Muse. If you followed our coverage of the Grok 4.5 launch or our full Grok 4.6 review, you will notice this is the first generation where xAI claims the top price-performance position so openly.
What is actually new in Grok 4.7
| Area | Grok 4.6 High | Grok 4.7 xHigh |
|---|---|---|
| Input price per million tokens | $2 | $2 |
| Output price per million tokens | $6 | $6 |
| CursorBench 4.0 | 40.4% | 46.3% |
| Terminal-Bench 4.0 | 20.3% | 38.0% |
| EEBench | 53.0% | 64.0% |
| AA Briefcase v1.1 | 1,546 | 1,657 |
| Harvey Legal Agent | 15.8% | 19.6% |
| HealthBench Professional | 48.5% | 56.7% |
Context: the jump comes from a larger new base model, longer reinforcement-learning training on a harder mix of tasks that take hours to finish, and native understanding of the Grok Bot environment. In plain language, the model was specifically trained to keep going on long, messy, realistic jobs instead of short exam-style questions.
That stamina shows up in AA Briefcase, which measures multi-hour office work. Grok 4.7 scores 1,657 there, ahead of Grok 4.6 at 1,546 and GPT-5.6 Sol Max at 1,487. For students and professionals who use AI to draft long reports or clean up large spreadsheets, endurance matters more than one clever answer.
While you decide whether to switch models, it helps to have the right workspace around you. Truescho's free all-in-one platform pairs AI study tools with scholarship listings, a GPA calculator, and 1500+ university rankings, so any new model you test has a real task waiting the same day.
Safety: the strongest guardrails xAI has shipped
Alongside the performance claims, xAI announced a completely new guardrail package for Grok 4.7. The company says it is the strongest model in its own refusal tests and jailbreak resistance so far.
On the LatchBio benchmark for biology safety, Grok 4.7 leads with 62.4%. On HackerBench v0.3, it complies with only 3.3% of dual-use hacking requests while rarely blocking legitimate security work. xAI is also giving invite-only access to cybersecurity partners so they can probe the model's red-team capabilities in depth. For security teams at banks, telcos, and universities worldwide, that combination is becoming a serious selection criterion rather than a footnote.
How to try Grok 4.7 today
Availability is broad from day one. Grok 4.7 is live in Cursor, Grok Build, the Grok API, external coding tools, cloud platforms, and routers, and the quickest way in is the free trial at x.ai/build.
Source: xAI docs
If you manage API budgets for a student project in Lagos, a startup in Manila, or a dev shop in Warsaw, the migration path is painless. Because pricing carries over from Grok 4.6, existing integrations keep working, and you can compare both generations side by side using the detailed numbers in our Grok 4.6 explained guide before committing tokens to the new model.
What this means for you
For developers everywhere from Karachi to São Paulo to Jakarta, the practical takeaway is simple: frontier-level coding help just got dramatically cheaper per token. At $6 per million output tokens, Grok 4.7 undercuts GPT-5.6 Sol Max by roughly five times on output cost and Fable 5.1 Max by more than eight times, while beating both on most of the coding and knowledge benchmarks published yesterday.
The announcement lists no regional restrictions on the API, so international users can sign up directly and experiment for free through Grok Build and Cursor before paying anything. Students preparing programming assignments can pair the model with structured study resources, and anyone comparing ecosystems can revisit our coverage of the GPT-5.6 launch to weigh OpenAI's offering against this release.
For security-conscious teams, the new guardrails matter as much as the raw scores. A coding model that refuses dangerous requests at high rates while still allowing legitimate security workflows is exactly the balance most enterprise buyers have been asking for all year.
Frequently asked questions
What is Grok 4.7 and when was it released?
It is SpaceXAI's most capable model for coding and knowledge work, announced on September 21, 2026. It runs longer on hard tasks, checks its work more carefully, and ships with the company's best-calibrated safety guardrails so far. It succeeds Grok 4.6 at the same price and speed.
How much does Grok 4.7 cost per million tokens?
Input tokens cost $2 per million and output tokens cost $6 per million, identical to Grok 4.6. A fast variant doubles the output speed and doubles the price. Compared with GPT-5.6 Sol Max at $4 and $20, and Fable 5.1 Max at $10 and $50, it is the cheapest premium option in its class.
Is Grok 4.7 better than Grok 4.6 for coding?
On the released benchmarks, yes. CursorBench 4.0 rises from 40.4% to 46.3%, Terminal-Bench 4.0 nearly doubles from 20.3% to 38.0%, and EEBench climbs from 53.0% to 64.0%. The gains come from a larger base model and longer reinforcement learning on multi-hour tasks.
How do I try Grok 4.7 in Cursor or Grok Build?
It is available today inside Cursor, Grok Build, the Grok API, external coding tools, cloud platforms, and routers. The quickest starting point is the free trial at x.ai/build, where you can test the model on your own tasks before connecting any paid API key.
How does Grok 4.7 compare with GPT-5.6 Sol and Fable 5.1?
It beats GPT-5.6 Sol Max on CursorBench, EEBench, Terminal-Bench, and Harvey Legal while charging about a third of its output price. Fable 5.1 Max leads several benchmarks, including CursorBench at 51.8%, but it charges $50 per million output tokens, more than eight times Grok's rate.
Does Grok 4.7 support the xAI API internationally?
The announcement lists no regional restrictions for the API, and developers worldwide can sign up directly. If you already use the Grok API with Grok 4.6, the new model arrives at the same pricing, so existing integrations in Cursor and other tools keep working without changes.
Yesterday's release quietly resets what a dollar buys in AI coding help. A model that tops several independent benchmarks at $2 and $6 per million tokens, with a free trial and no announced regional limits, removes the last excuse for not testing frontier models in your own workflow. Try it on one real task this week, compare it honestly with what you use now, and let the results decide your default. And while you experiment, keep Truescho's free platform open in the next tab for scholarships, GPA planning, and AI study tools that turn cheap AI power into finished coursework.
Sources
- xAI — Grok 4.7 announcement — official launch page with benchmarks, pricing, and safety details
- xAI docs — Grok 4.7 — official model documentation for developers
- Grok Build free trial — starting point to test Grok 4.7 at no cost