⚡ Quick verdict
- What it is → Grok 4.7 is xAI's flagship model, released 21 September 2026: a larger base model than Grok 4.6, trained longer on harder, multi-hour tasks, with a 500,000-token context window.
- The catch → On the independent Artificial Analysis Intelligence Index (v5.0.1, 15 November 2026) it scores 35.5 — far behind Claude Fable 5.2 at 64.6. It is a coding specialist, not the smartest generalist.
- Best for → Developers and X/Grok users who want strong coding benchmarks at $2 input / $6 output per million tokens — the cheapest flagship output price in its class.
- Skip it if → You want the best all-round model. Fable 5.2 and Gemini 4 Carbon score far higher overall, and Grok 4.7 even slips behind its own predecessor on FrontierMath.
- New this week → The November leaderboard reshuffle puts Grok 4.7 in context: Fable 5.2 is the new #1 on the AA Index, and Grok 4.7 sits around 16th of 54 models on that board.
We may earn a commission if you buy through our links, at no extra cost to you.
Grok 4.7 arrived quietly. No livestream, no keynote — xAI published a launch post and a model card on 21 September 2026, and the new flagship went straight into the API, Cursor and Grok Build. Two months later it is still the model xAI ships everywhere, and the November leaderboard reshuffle gives us the first honest picture of where it stands: strong where xAI said it would be strong, weak exactly where its critics expected.
This review is a researched comparison, not a hands-on test. It is built from xAI's own launch post and model card, xAI's API documentation, and independent measurements from Artificial Analysis, LMArena and Epoch AI, all checked in November 2026. Vendor benchmarks are labelled as vendor benchmarks throughout — xAI's numbers are one signal, not the verdict — and every price is dated November 2026, because xAI changed its consumer pricing in October and may change it again. If you want the head-to-head with its oldest rival first, our Grok vs ChatGPT comparison covers the two chatbots side by side, and our best AI model of November 2026 ranking puts Grok 4.7 against every flagship on one board.
What is Grok 4.7?
Grok 4.7 is the flagship model of xAI (now operating as SpaceX Super Intelligence, or SpaceXSI) — the model that powers the Grok chatbot on grok.com, inside the X app, and through the xAI API. It replaced Grok 4.6, which had shipped only six weeks earlier on 12 August 2026, and it is the model xAI recommends for both chat and coding in its developer docs as of November 2026.
xAI describes Grok 4.7 as "our most capable model for coding and knowledge work. It works longer on difficult tasks, checks its own work more carefully, and comes with our best-calibrated safeguards to date." The recipe, in the amount of detail xAI is willing to publish: a new, larger base model than Grok 4.6, a longer reinforcement-learning run over a harder mix of tasks weighted toward work that can take hours to complete, stronger self-verification, better long-context management, native training on the Grok Bot harness, and a new safeguard stack. Decrypt reports the model at about 2.1 trillion parameters, up from 1.5 trillion for Grok 4.6 — but xAI has not published an official parameter count, so treat that figure as reported, not confirmed.
The launch surfaces tell you who xAI built this model for. Day one was the API, Cursor, Grok Build, third-party coding harnesses, and model routers and cloud platforms — developers first. The model card said consumer surfaces (the web app, mobile apps and Grok inside X) would follow later; by November 2026, independent trackers report Grok 4.7 as the flagship model served in the Grok apps and on X, and Grok 4.6 remains available alongside it. There is also a Grok 4.7 Fast variant — the same model served on faster infrastructure — and as of 10 November 2026, no Grok 4.8 has shipped.
Grok 4.7 at a glance
The specification sheet is unusually stable for a flagship launch: almost everything matches Grok 4.6, which is the point — xAI sells this as a better model at the same price and speed, not a new product tier.
Grok 4.7 — key facts, November 2026
- Released
- 21 September 2026 (model ID grok-4.7)
- Context window
- 500,000 tokens
- Input / output
- Text and images in, text out
- Reasoning effort
- Low, medium, high (default), xhigh
- Knowledge cutoff
- May 2026 per xAI's model page (the model card cites a June 2026 pretraining cutoff, with supplemental data through August 2026)
- API price (<200K tokens)
- $2.00 input / $0.50 cached input / $6.00 output per million tokens
- API price (≥200K tokens)
- $4.00 input / $1.00 cached input / $12.00 output per million tokens
- Capabilities
- Function calling, structured outputs, web search, X search, code execution
- Batch API
- Not supported, per xAI's docs
- Consumer access
- grok.com, Grok iOS/Android apps, Grok inside X — via The X Subscription
Prices are xAI's listed US API prices as reported in November 2026 and change often — check the official site before buying.
Two details in that table deserve a second look. First, the context window: 500,000 tokens is large in absolute terms, but half the 1M window the other November flagships now share, and the price doubles once a prompt crosses 200,000 tokens — so the cheapest way to use Grok 4.7 is also the way that keeps prompts under that threshold. Second, the knowledge cutoff: xAI's own pages disagree by a month (May on the model page, June in the model card), a small inconsistency that tells you how fast this launch moved. Either way, Grok 4.7's training data ends in mid-2026, and for anything newer it leans on live web and X search — which, for a model built into a social network, is arguably the point.
Benchmarks: what xAI claims
xAI published a launch benchmark table comparing Grok 4.7 at xhigh effort against Grok 4.6, GPT-5.6 Sol Max and Claude Fable 5.1 Max. These are vendor figures — xAI's own numbers, run in harnesses xAI chose — and they should be read as "where xAI thinks it wins," not as independent proof. With that caveat, the pattern is consistent: Grok 4.7 beats Grok 4.6 on nearly every row, sometimes by a lot.
xAI launch benchmarks — vendor-reported, September 2026
- CursorBench 4.0
- Grok 4.7: 46.3% · Grok 4.6: 40.4% · GPT-5.6 Sol Max: 41.7% · Fable 5.1 Max: 51.8%
- DeepSWE v1.1
- Grok 4.7: 71.0% (high effort) · Grok 4.6: 65.2% · GPT-5.6 Sol Max: 72.7% · Fable 5.1 Max: 70.0%
- Terminal-Bench 4.0
- Grok 4.7: ~38% · Grok 4.6: 20.3% · Fable 5.1 Max: 57.9%
- SWE-Marathon v1.1
- Grok 4.7: 46.0% · Grok 4.6: 31.9%
- EEBench
- Grok 4.7: 64.0% (blog) / 66.0% (model card) · Grok 4.6: 53.0% · GPT-6 Astra: 69.3%
- HealthBench Professional
- Grok 4.7: 56.7% · Grok 4.6: 48.5% · Fable 5.1 Max: 62.1%
- Harvey Legal Agent Benchmark
- Grok 4.7: 19.6% · Grok 4.6: 15.8% · Fable 5.1 Max: 6.7%
- AA Briefcase v1.1
- Grok 4.7: 1,657 · Grok 4.6: 1,546 · Fable 5.1 Max: 1,678
All figures are xAI's launch-time claims, not independent tests. Competitor values reproduce xAI's own comparison table. Note the EEBench discrepancy: xAI's blog says 64.0%, its model card says 66.0%.
The strongest rows are the coding and agentic ones. Terminal-Bench 4.0 nearly doubles, from 20.3% to about 38% — the biggest single jump in the table, and the one that matches xAI's pitch of a model built for long, multi-hour work. SWE-Marathon jumps from 31.9% to 46.0%. DeepSWE reaches 71.0%, essentially tied with GPT-5.6 Sol Max at 72.7% and ahead of Fable 5.1 at 70.0%. And on the Harvey Legal Agent Benchmark, Grok 4.7's 19.6% is not just an improvement on 4.6's 15.8% — it is nearly triple Fable 5.1's 6.7%, the one row where Grok 4.7 leads xAI's own comparison table outright.
But the same table shows the ceiling. On CursorBench 4.0, Grok 4.7's 46.3% still trails Fable 5.1's 51.8%. On Terminal-Bench, 38% trails Fable 5.1's 57.9% by twenty points. On HealthBench Professional, 56.7% trails Fable 5.1's 62.1%. And two rows xAI did not headline are regressions: on FrontierMath, Grok 4.7 scores about 13 points lower than Grok 4.6, and it slips slightly on GPQA Diamond too. A model that gets better at long coding runs and worse at hard math is a model whose training mix changed — longer reinforcement learning on multi-hour tasks, at the cost of the skills that mix did not reward.
What independent tests say
Vendor tables flatter; independent boards do not. Three independent sources checked in November 2026 tell a consistent story: Grok 4.7 is a strong coding specialist with a mediocre overall score.
Artificial Analysis Intelligence Index. The most important independent number is the weakest one for Grok. On the AA Intelligence Index v5.0.1 — a composite of nine evaluations across knowledge work, coding, science and general knowledge — Grok 4.7 at xhigh scores 35.5 in the 15 November 2026 snapshot, against 64.6 for Claude Fable 5.2 and 60.0 for Gemini 4 Carbon. That puts Grok 4.7 around 16th of 54 models on that board. One caution when comparing with launch coverage: the index was re-versioned to v5.0.1 in November, so the 46 that some September reports quoted for Grok 4.7 was measured on the older version and is not comparable with the 35.5. On the current version, the gap to the leaders is large — roughly 29 points behind Fable 5.2 — and it is not a rounding error. Artificial Analysis also reports Grok 4.7's hallucination rate at 29%, down from 34% for Grok 4.6: better, but still nearly one answer in three.
LMArena. On the human-preference board, Grok 4.7 scored 1,444 on text as of 15 October 2026 — good enough for the top 100, not the top 10 — and 1,640 on WebDev, where it looks considerably stronger. Preference boards measure what users like, not what is correct, and new models take time to settle; still, the split between a middling text score and a strong WebDev score fits the specialist picture.
Epoch AI and writing tests. An independent Epoch AI run puts Grok 4.7 at 92.7% on GPQA Diamond and 98.1% on the OTIS Mock AIME — strong results that sit oddly next to the FrontierMath regression, and a reminder that single benchmarks swing. On writing, ToneBench ranks Grok 4.7 28th of 87 models through its best setting (medium), with a writing score of 79.6 out of 100 — respectable, not leading. And on ARC-AGI-3, the new visual-reasoning benchmark, the published semi-private snapshot puts Grok 4.7 at 1.8% against Fable 5.2's 81.2% — a rout, though ARC-AGI-3 is a young benchmark and its scores are unusually spread out.
The honest synthesis: if your work is coding, long agentic runs, or legal-adjacent knowledge work, Grok 4.7's independent numbers are competitive with models that cost several times more. If your work is general knowledge, math, or writing, the independent boards say to look elsewhere. That split is the whole review in one paragraph, and it is why the price matters so much — which brings us to the money.
Grok 4.7 vs the rivals
Here is where Grok 4.7 sits against the flagships most readers will actually choose between, on the November 2026 independent index and on price. The comparison is deliberately narrow: overall capability against output price, because that is the trade-off Grok 4.7 forces you to make.
Flagship comparison — November 2026
- Grok 4.7 (xAI)
- AA Index 35.5 · $2 in / $6 out per MTok · 500K context · Best for: X ecosystem, long coding runs
- Claude Fable 5.2 (Anthropic)
- AA Index 64.6 · $8 in / $40 out per MTok · 1M context · Best for: overall capability, coding, agents
- Gemini 4 Carbon (Google)
- AA Index 60.0 · price not yet the value story · 1M context · Best for: overall capability at Google scale
- GPT-6.1 Sol (OpenAI)
- AA Index 51–55 (mid-tier range) · $1–2 in per MTok · Best for: everyday work at low cost
- DeepSeek (open-weight)
- Strong coding per dollar · fraction of flagship prices · Best for: budget coding and self-hosting
AA Index figures are Artificial Analysis Intelligence Index v5.0.1, snapshot 15 November 2026. Prices are listed US API prices as reported in November 2026 and change often — check the official site before buying.
The price gap is the story. Grok 4.7's $6 per million output tokens is less than a sixth of Fable 5.2's $40 — and on DeepSWE, Grok 4.7 (71.0%) is essentially tied with the models that cost five times more. If you are running thousands of coding-agent calls a day, that arithmetic is hard to ignore: you can afford roughly six Grok 4.7 outputs for every Fable 5.2 output, and on the coding benchmarks that matter to you, the cheaper model is not six times worse. It is, on some boards, barely worse at all.
The counter-argument is just as real. Fable 5.2's 64.6 on the AA Index is not a small lead — it is nearly double Grok 4.7's 35.5 — and it shows up exactly where Grok 4.7 is weakest: general knowledge, science, and reasoning breadth. A coding agent that writes good code but misunderstands the task is not a bargain at any price. And GPT-6.1 Sol, in the 51–55 AA range at $1–2 per million input tokens, undercuts Grok 4.7 on price while scoring higher overall — which makes Grok 4.7's value case narrower than xAI's marketing suggests. For the full field, see our best AI model of November 2026 ranking and our Gemini 4 Hydrogen vs Claude Fable 5.2 comparison. If you are choosing between chatbots rather than APIs, our ChatGPT vs Gemini comparison and our Grok vs ChatGPT head-to-head cover the consumer side.
Try Claude → Try ChatGPT → Try Gemini → Try DeepSeek →
Pricing: the API vs The X Subscription
Grok has two price lists, and confusing them is the most common mistake in Grok pricing coverage. There is the API price for developers, and the consumer subscription for chat — and since 12 October 2026, the consumer side has a new name.
The API. Grok 4.7 costs $2.00 per million input tokens, $0.50 per million cached input tokens and $6.00 per million output tokens for prompts under 200,000 tokens. Cross that 200K threshold and the whole request bills at the higher band: $4.00 input, $1.00 cached, $12.00 output. The long-context surcharge is the detail most coverage skips, and it matters: a 500K context window that doubles in price past 200K is not the same product as a flat-rate 500K window. xAI also lists rate limits of 150 requests per second and 50,000,000 tokens per minute, and marks Grok 4.7 as unsupported by the Batch API — so there is no discounted batch route for bulk work. Grok Voice 3.0, the speech-to-speech API, is billed separately at $0.08 per minute.
The consumer subscription. On 12 October 2026, xAI retired the SuperGrok tiers and replaced them with The X Subscription — one bill covering Grok, Grok Bot, Cursor and X Premium, with a single weekly usage pool shared across them. The tiers, as checked against xAI's pricing on 5 November 2026: Free ($0), Plus ($8 a month), Premium ($30), Super ($100) and Ultra ($200). Premium absorbs the old SuperGrok, Cursor Pro and X Premium+ benefits, which xAI says cost $90 a month combined. Existing SuperGrok, Cursor and X Premium subscribers can keep their old plans — nobody is moved automatically, and switching to the new bundle is permanent, according to xAI's own docs. Usage scales 1× / 4× / 15× / 50× against Plus across the paid tiers, and the pool resets weekly; unused usage does not carry over, according to user reports. All of this is third-party-tracked pricing, not a live checkout — prices change often, so confirm on x.ai before subscribing.
The X Subscription — November 2026
- Free
- $0 — daily, limited Grok access in the web and mobile apps, starter Cursor allowance
- Plus
- $8/month — the cheapest paid entry, biggest step up from Free for regular users
- Premium
- $30/month — 4× Plus usage, Grok Bot, cloud agents, X Premium features
- Super
- $100/month — 15× Plus usage, priority at peak times
- Ultra
- $200/month — 50× Plus usage, earliest access, fastest generations
Consumer prices as checked against x.ai/pricing on 5 November 2026 by independent trackers. Legacy SuperGrok and X Premium+ subscribers may keep old plans. Prices change often — check the official site before subscribing.
Which tier actually buys you Grok 4.7? The honest answer is that xAI does not publish a clean per-model, per-tier table, and coverage disagrees on the details. What is consistent across November 2026 trackers: Grok 4.7 is the flagship model served in the Grok apps and on X, the free tier gives daily limited access, and higher tiers buy more usage, faster generations and earlier access rather than a different model. Do not pay for a tier on the assumption that it unlocks a model the free tier hides — pay for usage limits and speed, and verify the current tier table on x.ai before you subscribe.
How to access Grok 4.7
There are four ways in, and they serve different people. Chat: open grok.com in a browser, sign in with X, Google, Apple or email, and pick Grok 4 from the model picker; or open Grok inside the X app, or use the standalone Grok apps on iOS and Android. API: use the model ID grok-4.7 in the xAI API, with the reasoning_effort parameter set to low, medium, high (the default) or xhigh — xhigh is the setting behind xAI's best benchmark numbers, and the one that costs the most in output tokens. Coding tools: Grok 4.7 shipped day-one in Cursor and Grok Build, and through model routers and cloud platforms. Grok Bot: the X-native agent harness Grok 4.7 was trained to understand natively — the model and the bot were built together, which is why xAI describes Grok 4.7 as an agentic model rather than a chat model with tools bolted on.
One practical note from early adopters: the API is not the same experience as the app. Independent coverage in the launch weeks reported usage-balance walls on API accounts — the model list says 4.7 is available and default, but a call can still hit a balance limit before it completes. If you are building on the API, fund the account and test a real xhigh call before you commit a workflow to it. And if you are choosing a coding assistant rather than a raw model, our best AI coding assistants roundup compares the tools that wrap these models, and our best AI SEO tools roundup covers the research side of the same workflow.
Grok 4.7 vs Grok 4.6: should you switch?
Yes — with one exception. Grok 4.7 costs exactly what Grok 4.6 costs, runs at the same speed, has the same 500K context window, and beats 4.6 on nearly every benchmark xAI publishes. There is no price or compatibility reason to stay on 4.6: the model ID changes, the bill does not. xAI itself calls 4.7 "a notable improvement over Grok 4.6 at the same price and speed," and on the numbers that is fair — CursorBench up 5.9 points, DeepSWE up 5.8, Terminal-Bench nearly doubled, SWE-Marathon up 14.1 points, HealthBench up 8.2 points.
The exception is math-heavy work. Grok 4.7's FrontierMath regression — about 13 points below Grok 4.6 — is the one place the new model is measurably worse than the old one, and the GPQA Diamond slip points the same way. If your workload is mathematical reasoning, scientific computing or benchmark-style problem solving, test Grok 4.7 against 4.6 on your own tasks before you migrate — and consider that the independent Epoch AI run (92.7% on GPQA Diamond) disagrees with the direction of xAI's own GPQA numbers, which means the truth here is genuinely unsettled. For everyone else — coding, long agentic runs, knowledge work, legal-adjacent tasks — switching is a free upgrade.
Who Grok 4.7 is for — and who should skip it
Grok 4.7 is a good fit if: you already live in X and the Grok apps, where it is the native flagship and needs no new account or workflow; you are a developer running coding agents at volume, where $2/$6 per million tokens buys near-top-tier DeepSWE and CursorBench scores; you run long, multi-hour agentic tasks and need a model trained for them; or you want the cheapest flagship output price in its class and are willing to trade general-knowledge breadth for it.
Grok 4.7 is not a good fit if: you want the smartest general-purpose model — Fable 5.2 (64.6) and Gemini 4 Carbon (60.0) are in a different league on the independent index; your work is math or science-heavy, where Grok 4.7 regressed against its own predecessor; you need a 1M context window without a price doubling past 200K tokens; or you want a model with a long independent track record — Grok 4.7 is eight weeks old, and the independent evidence is still thin. If you are in the "best model, price aside" camp, our November 2026 model ranking is the better starting point; if you want the cheapest capable chatbot, our free-tier chatbot comparison covers what $0 actually buys.
Grok 4.7
Strengths
- Strong coding benchmarks at the cheapest flagship output price: 71.0% DeepSWE, 46.3% CursorBench 4.0, $6 per million output tokens.
- Terminal-Bench 4.0 nearly doubled vs Grok 4.6 (about 38% vs 20.3%) — real progress on long agentic runs.
- Leads xAI's own table on the Harvey Legal Agent Benchmark (19.6% vs Fable 5.1's 6.7%).
- 500K context window with four reasoning effort levels, default high, xhigh for the hardest tasks.
- Native to X and the Grok apps — no new account for X users — plus day-one Cursor and Grok Build support.
- Free tier with daily access, and a hallucination rate down to 29% from 34% on Grok 4.6.
Weaknesses
- Weakest overall independent score of the November flagships: 35.5 on the AA Intelligence Index v5.0.1, vs 64.6 for Fable 5.2.
- Math regression: about 13 points worse than Grok 4.6 on FrontierMath, slight slip on GPQA Diamond.
- 500K context is half the 1M rivals offer, and the API price doubles past 200K tokens.
- ARC-AGI-3 semi-private snapshot: 1.8% vs Fable 5.2's 81.2% — a big visual-reasoning gap.
- No Batch API support, and xAI's own pages disagree on small facts (EEBench 64% vs 66%, cutoff May vs June).
- Eight weeks old with thin independent evidence — most claims still rest on xAI's own material.
What to watch before you trust it
Four things are worth watching as Grok 4.7 matures. First, safeguards. xAI calls this its best-calibrated safeguard stack to date, and on its own HackerBench test it reports Grok 4.7 letting through 3.3% of risky dual-use prompts while rarely blocking legitimate security work. That is xAI grading its own homework — independent safety evaluations of Grok 4.7 had not been published in the sources we checked — so treat the safeguard claim as "vendor-reported, unverified" rather than settled. Grok's history of controversial outputs makes this the claim to verify, not the one to assume.
Second, the price structure. The $2/$6 headline is real, but the 200K-token threshold doubles it, and the consumer side changed entirely in October when SuperGrok was retired for The X Subscription. Anyone quoting a Grok price from before 12 October 2026 is quoting a dead product. Check the live price on x.ai before you build a budget on either number.
Third, the documentation gaps. A model whose blog says 64.0% and whose model card says 66.0% on the same benchmark, and whose cutoff is May on one page and June on another, is a model whose documentation is moving faster than its reviewers. Small discrepancies do not make the big numbers wrong, but they are a reason to verify any single figure that matters to you against the live docs rather than a launch post.
Fourth, the generalist gap. Grok 4.7's coding scores say "flagship"; its AA Index score says "mid-tier." Both can be true at once, and the gap is the decision: if your work matches Grok 4.7's strengths, it is a bargain; if it matches its weaknesses, no price makes it the right model. The next independent index update — and whether xAI ships Grok 4.8 before it — will tell us whether the gap is closing or structural.
Our verdict
Grok 4.7 is the most honest deal in the flagship market and the most lopsided one. On coding and long agentic work it delivers near-top-tier benchmark scores at $2 input and $6 output per million tokens — a price no other flagship touches — inside the apps and the social network millions of people already use. That is a real, specific, valuable thing, and for developers running coding agents at volume it is probably the best value in AI right now.
But the overall picture is weaker than the coding picture, and the independent boards are blunt about it: 35.5 on the AA Intelligence Index against 64.6 for Claude Fable 5.2, a math regression against its own predecessor, a 500K window that costs double past 200K tokens, and a safety story that still rests on xAI's own word. Grok 4.7 is not the best model of November 2026. It is the best model for a specific job at a specific price — and if that job is yours, the free tier is sitting there waiting for you to test it.
Bottom line: a coding specialist at a bargain price, not a generalist champion. Try it free on grok.com, run it on your real work, and only pay once it has earned the subscription. Check the official site for current pricing and availability before you pay.
Frequently asked questions
What is Grok 4.7?
Grok 4.7 is xAI's flagship AI model, released on 21 September 2026. It is built on a larger base model than Grok 4.6, with a longer reinforcement-learning run weighted toward tasks that take hours, a 500,000-token context window, and four reasoning effort levels (low, medium, high and xhigh). It accepts text and image input and returns text, and it is the model served in the Grok apps, on X, and through the xAI API.
How much does Grok 4.7 cost?
On the xAI API, Grok 4.7 costs $2.00 per million input tokens, $0.50 per million cached input tokens and $6.00 per million output tokens for prompts under 200,000 tokens; longer prompts move to a higher band of $4.00 / $1.00 / $12.00. For consumers, Grok is sold through The X Subscription: Free ($0), Plus ($8), Premium ($30), Super ($100) and Ultra ($200) a month. Prices change often — check the official site before you pay.
Is Grok 4.7 free to use?
Yes, with limits. Grok has a free tier on X and in the Grok apps with daily, limited access — you can chat with Grok 4.7 without paying, but heavy use, higher limits, voice mode, Imagine image and video generation, and the earliest access to new models sit behind the paid tiers of The X Subscription, starting at Plus ($8 a month). The free tier is enough to try the model; it is not a replacement for a paid plan if you use Grok all day.
How does Grok 4.7 compare to Grok 4.6?
Grok 4.7 is a straight upgrade on coding and long-running work at the same price and context window: xAI reports CursorBench 4.0 up from 40.4% to 46.3%, DeepSWE up from 65.2% to 71.0%, Terminal-Bench 4.0 up from 20.3% to about 38%, and SWE-Marathon up from 31.9% to 46.0%. The price is identical ($2 input / $6 output per million tokens) and the 500K context window is unchanged. The one regression xAI's own numbers show is math: Grok 4.7 scores about 13 points lower than Grok 4.6 on FrontierMath and slightly lower on GPQA Diamond.
How does Grok 4.7 rank against Claude Fable 5.2 and other flagships?
Behind the leaders on overall capability, ahead on price. On Artificial Analysis' Intelligence Index v5.0.1 (snapshot 15 November 2026), Grok 4.7 at xhigh scores 35.5, far behind Claude Fable 5.2 at 64.6 and Gemini 4 Carbon at 60.0 — but Grok 4.7's $6 per million output tokens is a fraction of Fable 5.2's $40. On specific coding benchmarks Grok 4.7 looks much stronger: xAI reports 71.0% on DeepSWE and 46.3% on CursorBench 4.0, close to or ahead of several pricier rivals. It is a specialist, not the best generalist.
Is Grok 4.7 worth it?
It is worth it if you already live in the X and Grok ecosystem, or if you are a developer who wants strong coding benchmarks at $2/$6 per million tokens — the cheapest flagship output price in its class. It is not worth switching to if you want the smartest general-purpose model: Claude Fable 5.2 and Gemini 4 Carbon score far higher on the independent Artificial Analysis index, and Grok 4.7's math and general-knowledge scores trail its own predecessor in places. Try the free tier first, and only pay once Grok 4.7 has proved itself on your actual work.