Grok 4.7
A model by xAI · released 2026-09-21 · 40 checked facts from 9 sources
Timeline
- xAI released Grok 4.7, Sep 21, 2026
- Grok 4.7 is available on Cursor
- Grok 4.7 is available on Grok Build
- Grok 4.7 is available on xAI API
- SpaceX Super Intelligence released Grok 4.7, September 21, 2026
Prices
Per price, each value with the dates it was seen. A value seen on one date only is one copy of the page.
No price has changed over the record.
| Price for | Values over time |
|---|---|
| per 1M input tokens | $2 |
| per 1M output tokens | $6 |
Scores
Self-reported: stated on the lab's own channel. A score reported twice with two values is shown twice, with
its date: a lab re-runs a benchmark under a new setup; neither replaces the other.
| Benchmark | Score | Date | Who |
|---|---|---|---|
| AA-Briefcase | 1638 Elo | 2026-10-29 | reported by artificialanalysis.ai |
| APEX-1 + APEX-Agents | 58.1% | 2026-10-16 | reported by mercor.com |
| APEX-1 + APEX-Agents | 58.7% | 2026-10-16 | reported by mercor.com |
| Design Creativity Bench | 62.2 | 2026-10-14 | reported by arcfu.com |
| LatchBio's biosafety benchmark | 62.4% | 2026-09-21 | self-reported |
| Models Engineering Index | 41.7 | 2026-11-19 | reported by git-aarvin.github.io |
What it does
- Grok 4.7: risky dual-use prompts allowed through 3.3%
- Grok 4.7: most capable model for coding and knowledge work
- Grok 4.7: works longer on difficult tasks
- Grok 4.7: checks its own work more carefully
- Grok 4.7: comes with our best-calibrated safeguards to date
- Grok 4.7: better at verifying its own work
- Grok 4.7: managing longer context
- Grok 4.7: natively understand the Grok Bot harness
- Grok 4.7: better at conversational tasks and general knowledge work
- Grok 4.7: better at creating documents and presentations
- Grok 4.7: strongest model on refusals and jailbreak resistance
- Grok 4.7: leads on utility for benign tasks and safe refusal on dangerous ones
- Grok 4.7: balances strong cyber defense capabilities with low refusal rates for legitimate use
- Grok 4.7: highest safety on HackerBench v0.3
- Grok 4.7: uses a new, larger base model compared to Grok 4.6
- Grok 4.7: trained with a longer reinforcement learning run on a harder mix of tasks
- Grok 4.7: 500k context window
- Grok 4.7: works longer on difficult tasks and checks its own work more carefully
- Grok 4.7: agentic tool calling
- Grok 4.7: configurable reasoning
- Grok 4.7: minimal hallucinations
- Grok 4.7: messages sent in 15 simulated days 187
- Grok 4.7: emails sent to government officials, council members and the press 69
- Grok 4.7: messages sent in first week 116
- Grok 4.7: share of answers recommending the Republican candidate 92%
- Grok 4.7: share of answers citing Turning Point Action’s site 23%
- Grok 4.7: share of answers citing X 41%
In the news
- git-aarvin.github.io: Gemini 4 Neon is the new #1 on the Models Engineering Index
- stephenjgallagher.substack.com: Claude Opus 5.5 Can Draw Permit-Level Architectural Drawings
- artificialanalysis.ai: Gemini 4 Carbon is the new #1 on the Artificial Analysis Intelligence Index v5
- TechCrunch: Forced to pick, Grok’s AI model chooses Republicans. OpenAI’s picks Democrats.
- vals.ai: How Far Will AI Assistants Go for a Refund They Can't Get?
- mercor.com: 90% of US working hours are spent on tasks not covered well by AI benchmarks
- arcfu.com: Why do LLM pelicans look like this?
- Simon Willison: Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and a new price war
Built 2026-11-21 06:50 UTC from the news
knowledge graph by code. Every line is a fact whose quote was checked against the source's own words (hover a source);
it says who states it, not that it is true. Archived pages are the lab's pages as the Wayback Machine holds them.