Grok 4.7 is xAI's newest flagship model, released on September 21, 2026, forty days after Grok 4.6. xAI pitches it as its most powerful model yet for coding and knowledge work, built on a new, larger base model and served at the same price and speed as its predecessor. If you are coming from Grok 4.5, it is two generations newer.
This Grok 4.7 review covers what changed, what xAI's benchmark table and independent tests show, what it costs, and where its limits sit. It also covers access: Grok 4.7 is not yet integrated into Chat Smith, but Grok 4.5 is already live there, so you can use xAI's model line today without a separate subscription.
Grok 4.7 is the successor to Grok 4.6 and is now the model xAI recommends for most work, including code. Unlike Grok 4.6, which was mainly a post-training update, Grok 4.7 is built on a new, larger base model, followed by a longer reinforcement learning run on a harder mix of tasks weighted toward problems that take many hours to complete.
On the surface, little else changes. Grok 4.7 keeps a 500,000-token context window, text and image input, text output, and reasoning levels from low to xhigh, with a knowledge cutoff of May 2026. xAI says it works longer on difficult tasks, checks its own work more carefully, and is trained to understand its Grok Bot agent harness natively.
It is also part of a fast release cycle: Grok 4.5 in July, Grok 4.6 in August, and Grok 4.7 in September. For where each version sits, see our full list of Grok models.
Four areas show where Grok 4.7 moves past Grok 4.6.
On xAI's own table, Grok 4.7 at xhigh effort scores 46.3% on CursorBench 4.0, up from 40.4% for Grok 4.6, and 71.0% on DeepSWE v1.1 at high effort, a gain of nearly 6 points. The biggest jump is on Terminal-Bench 4.0, a test of multi-hour terminal work, where it scores 38.0% against Grok 4.6's 20.3%. On VulcanBench's independent Frontier v4 coding leaderboard, Grok 4.7 run through Cursor took the top three spots in early October, though the board notes those scores are not strictly comparable with its other columns.
Grok 4.7 also improves on professional tasks. It scores 64.0% on EEBench, an electrical engineering test, up from 53.0% and ahead of Claude Fable 5.1 Max, and 19.6% on the Harvey legal agent benchmark, also ahead of Fable 5.1. On AA-Briefcase v1.1, which simulates multi-hour office work, it reaches an Elo of 1,657 against 1,546 for Grok 4.6. For technical study and problem-solving, such as working through circuits with an AI physics solver, the engineering gains are the most relevant.
On Artificial Analysis's new Cyber Index, a defensive-security benchmark covering finding, validating, and patching vulnerabilities, Grok 4.7 at xhigh tied for first place at 56%. Part of that lead reflects willingness as well as skill: several OpenAI and Anthropic flagships refused at least 98% of the tasks, which made their performance hard to measure.
Artificial Analysis measured Grok 4.7's hallucination rate at 29%, down from 34% for Grok 4.6, and scored it about 46 on its Intelligence Index, ranking it 16th of 655 models at launch. The catch is that it thinks much longer to get there, using roughly 81,000 output tokens per index task against 36,000 for Grok 4.6. For heavy document work, such as asking it to summarize a long PDF, that extra effort shows up as longer waits.
The table below uses xAI's launch figures for Grok 4.7 at xhigh effort, Grok 4.6 at high, and Claude Fable 5.1 at max, the strongest rival in xAI's comparison. Prices are API rates per million tokens.
| Grok 4.7 | Grok 4.6 | Claude Fable 5.1 | |
|---|---|---|---|
| Input / output price | $2 / $6 | $2 / $6 | $10 / $50 |
| CursorBench 4.0 | 46.3% | 40.4% | 51.8% |
| Terminal-Bench 4.0 | 38.0% | 20.3% | 57.9% |
| EEBench (electrical engineering) | 64.0% | 53.0% | 56.4% |
| AA-Briefcase v1.1 (Elo) | 1,657 | 1,546 | 1,678 |
| Harvey Legal Agent Benchmark | 19.6% | 15.8% | 6.7% |
| HealthBench Professional | 56.7% | 48.5% | 62.1% |
| Available in Chat Smith | Not yet | Not yet | — |
The pattern is clear. Grok 4.7 beats Grok 4.6 on every row at the same price, and it leads Claude Fable 5.1 on engineering and legal tasks. Fable still leads on the main coding and long terminal benchmarks, but at five times the input price and more than eight times the output price.
Grok 4.7 keeps Grok 4.6's price: $2 per million input tokens, $0.50 for cached input, and $6 per million output tokens. Prompts of 200K tokens or more are billed at $4, $1, and $12 across the whole request. A Fast variant, available only in Cursor and Grok Build, runs at twice the output speed for twice the price. Because Grok 4.7 uses more reasoning tokens per task, the same job can cost noticeably more than on Grok 4.6, even at identical rates.
Grok 4.7 is competitive, not dominant. xAI itself calls it highly competitive in its class, and on its own table Claude Fable 5.1 Max still leads on CursorBench, Terminal-Bench, HealthBench, and GDPval. Independent coding results also disagree: VulcanBench ranks it first, while Snorkel AI's Senior SWE-Bench placed it sixth, with only a small gain over Grok 4.6.
It is also slow and token-hungry on hard tasks. At xhigh effort, VulcanBench measured about 28 minutes per coding task, and Artificial Analysis found it uses more than twice the output tokens of Grok 4.6 per task. That makes the low list price less of an advantage than it looks for heavy reasoning work.
The 500K context window is smaller than the 1M windows many rivals offer, and the doubled rate starts at 200K. Most published benchmark figures are also xAI's own, so test Grok 4.7 against alternatives such as Claude Sonnet 5 on your real tasks before committing.
Grok 4.7 is not integrated into Chat Smith yet, but xAI's model line already is. Grok 4.5 is live in Chat Smith on Android, iOS, and the browser — pick it from the model list and start chatting, with no xAI account or Grok subscription needed.
Grok 4.5 shares Grok 4.7's 500K context window and price tier, and it is far lighter on reasoning, so it starts answering quickly and suits everyday questions, writing, research, and quick coding help. Grok 4.7's biggest gains show up mainly in multi-hour agentic coding and specialist professional tasks that most chat users rarely run.
Chat Smith also lets you compare Grok 4.5 with DeepSeek V4 Pro and other leading models in the same app, so you can put one question to several models and keep the answer that holds up. Chat Smith Pro adds higher usage limits and more premium models for $9.99 per month or $39.99 per year.
Grok 4.7 is xAI's strongest model yet: a larger base model, clear gains over Grok 4.6 on every benchmark xAI published, standout results in engineering, legal, and defensive security work, and the same $2/$6 price. Its weak spots are long thinking times, high token use per task, a 500K context window, and a coding lead over rivals that depends on which benchmark you trust.
Until Grok 4.7 reaches more apps, Grok 4.5 remains an easy way to use xAI's models every day. You can try it now in Chat Smith, an AI chat app that puts it next to the latest models from OpenAI, Anthropic, Google, and DeepSeek.
Grok 4.7 is xAI's flagship model, released on September 21, 2026, built on a new, larger base model and trained for long, multi-hour tasks. It beats Grok 4.6 on every benchmark in xAI's launch table at the same API price of $2 per million input tokens and $6 per million output tokens. Chat Smith has not integrated Grok 4.7 yet, but you can use Grok 4.5 there alongside other leading AI models in one app.
The Chat Smith Editorial Team brings together AI enthusiasts, researchers, and creators to make AI accessible and practical. We share trends, tool reviews, and actionable guides that help individuals and businesses stay informed, boost productivity, and get more from AI.
