Grok 4.7

What is Grok 4.7?#

Grok 4.7 is xAI's top language model, released September 21, 2026 for coding, long-running agents, and knowledge work. It accepts text and image input, outputs text, and runs on Unifically as xai/grok-4.7. The context window is 500k tokens, the knowledge cutoff is May 2026, and reasoning effort is tunable across four levels: low, medium, high (the default), and xhigh. Unlike Grok 4.6, which was Grok 4.5 trained further, Grok 4.7 sits on a new, larger base model with a longer reinforcement learning run weighted toward tasks that take hours.

What's new in Grok 4.7#

Opus-level CursorBench 4.0 score at half the cost

46.3% on CursorBench 4.0 at Extra High effort, next to Claude Opus 5 Max at 46.6% and ahead of GPT-5.6 Sol Max at 41.7%. Cursor's board puts the average cost at $6.01 per task against $11.95 for Opus 5 Max. Claude Fable 5.1 Max still leads at 51.8%, at $17.28 per task.

Top five on agentic legal work

Vals AI ranks it fifth of 63 models on Harvey's Legal Agent Benchmark at 19.58%, up from 15.8% for Grok 4.6, and fifth of 37 on Public Benefits Bench at 68.54%. These are multi-step tasks over real case files, not quiz questions.

Built for tasks that run for hours

A new, larger base model and a longer reinforcement learning run on harder, many-hour problems. It checks its own work more often and manages long context better, which shows on Terminal-Bench 2.1: 76.03% on Vals AI, 14th of 67.

A small step up in general intelligence

Artificial Analysis scores it 46 on Intelligence Index v4.3.2, 16th of 655 models, against 44 for Grok 4.6. Claude Fable 5.1 and GPT-6 Astra lead that board at 53.

Four reasoning efforts on one model

low, medium, high, and xhigh trade speed and token spend against depth per request. On CursorBench 4.0 that is 33.1% at $1.58 per task at low and 46.3% at $6.01 at xhigh.

Best for#

Use cases#

Use Grok 4.7 behind coding agents that take a real issue from report to green tests: read the repo, patch, run, and try again. It fits legal and compliance assistants that read case files, apply rules, and return structured JSON through function calling. Office agents that draft reports, spreadsheets, and slide decks are a stated training focus. With 500k tokens of context, long contracts, design docs, and issue threads fit in one request, with screenshots included as image input.

Limitations#

Grok 4.7 is slow and wordy on long tasks. Artificial Analysis counted 240M output tokens across its Intelligence Index run against a 92M median, and Vals AI measured about 27 minutes per test on the Vals Index. Token spend climbs on the longest agent runs: Vals AI logged $12.73 per test on Terminal-Bench 4.0, where it scored 12.12%, 16th of 29.

It is not a clean win over Grok 4.6 everywhere. On the Vals Index it scores 54.15%, 24th of 59, behind Grok 4.6 at 59.17%. Competition math and programming are weak spots: 39.39% on IOI and 34.00% on ProofBench v1.1.

The 500k context window is half the 1M tokens Grok 4.3 takes, which matters for the largest single-request workloads. Output is text only, and image generation is a separate Grok Imagine family.

Grok 4.7 vs Grok 4.6#

Grok 4.7 is a new, larger base model, where Grok 4.6 was Grok 4.5 trained further. The gains are in long agent work: CursorBench 4.0 rises from 40.4% to 46.3%, Harvey's Legal Agent Benchmark from 15.8% to 19.6%, and the Artificial Analysis Intelligence Index from 44 to 46. The broad Vals Index goes the other way, 54.15% against 59.17%. Input and output tokens cost the same on both. Move coding-agent, legal, and office traffic to 4.7, and keep Grok 4.6 where your prompts are already tuned and scoring well.

When to use Grok 4.7#

Use Grok 4.7 when the task is long and the finish line matters: multi-hour coding runs, case-file review, report and deck building, and tool-calling workflows that punish shallow answers. Skip it when you need fast first tokens or the cheapest possible bulk calls.

API examples#

Call Grok 4.7 from any language by POSTing to /v1/chat/completions, the OpenAI-compatible endpoint shared by every language model on the platform. Full parameter docs live at docs.unifically.com/models/llm/xai/grok-4.7.

curl -X POST https://api.unifically.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "xai/grok-4.7",
    "messages": [
      { "role": "user", "content": "Read this service agreement and list every clause that shifts liability to the customer." }
    ]
  }'

The response comes back synchronously with the completion. Set "stream": true to receive tokens as they generate.