Grok 4.7 is SpaceXAI's AI model, released 21/09/2026. xAI's model card presents Grok 4.7 as its most capable model for long-running coding, engineering and office work, with large gains over Grok 4.6 on Terminal-Bench 4.0, EEBench and the Legal Agent Benchmark. It still trails Claude Fable 5.1 on most coding rows and GPT-6 Astra on EEBench and HealthBench Professional. The card documents safety evaluations against xAI's own Frontier Artificial Intelligence Framework.
Visit Grok 4.7AI-Powered
Leverages advanced AI technology to deliver cutting-edge capabilities and results.
Fast & Efficient
Optimized performance ensures quick results without compromising on quality.
Purpose-Built
Specifically designed for llms tasks and workflows.
How does Grok 4.7 score on independent benchmarks?
Grok 4.7 ranks #16 of 28 models on the Artificial Analysis Intelligence Index (v5.0.1) with a score of 35.5, as measured on 22 November 2026.
Where does Grok 4.7 rank against other models?
How good is Grok 4.7 at agentic coding?
How does Grok 4.7 do on individual tests?
| Evaluation (run by Artificial Analysis) | Grok 4.7 | Best of any model |
|---|---|---|
| Terminal-Bench 4.0 | 25.8% | 67.2% |
| FrontierCode 2.0 | 28.7% | 48.8% |
| Humanity's Last Exam | 43.1% | 74.3% |
| GDPval-AA (Elo) | 1717 | 1866 |
How has Grok 4.7's score changed over time?
We have tracked Grok 4.7 on Intelligence Index v5.0.1 since 22/11/2026. A point is added each day the score, coding index or price changes, so this chart fills in over time.
What does the Grok 4.7 system card report?
xAI's model card presents Grok 4.7 as its most capable model for long-running coding, engineering and office work, with large gains over Grok 4.6 on Terminal-Bench 4.0, EEBench and the Legal Agent Benchmark. It still trails Claude Fable 5.1 on most coding rows and GPT-6 Astra on EEBench and HealthBench Professional. The card documents safety evaluations against xAI's own Frontier Artificial Intelligence Framework.
These are xAI (SpaceXAI)'s own published results (self-reported, with the vendor's harness and settings), so they are not directly comparable with other labs' numbers. Sources: Grok 4.7 Model Card (21/09/2026), Introducing Grok 4.7 (21/09/2026), Grok models and pricing (xAI docs). Last checked 22/11/2026.
Coding
| Benchmark | Grok 4.7 | GPT-5.6 Sol | Claude Fable 5 (max, with fallback) | Grok 4.6 |
|---|---|---|---|---|
| CursorBench 4.0xhigh; Cursor long-horizon sessions | 46.3% | n/a | n/a | n/a |
| CursorBench 4.0 (high)high effort | 43.9% | n/a | n/a | n/a |
| DeepSWE v1.1high effort; mini-SWE-agent harness; Datacurve; pass@1 | 71% | 72.7% | 69.7% | 65.2% |
| Terminal-Bench 4.0xhigh; Grok Build harness; Harbor; the launch post table prints 37.6% | 38% | 37.3% | n/a | 20.3% |
| FrontierSWE V2xhigh; mean@5 partial-credit reward; Proximal Labs | 29% | 32.2% | n/a | 25.3% |
| SWE-Marathon v1.1high; resolution rate; Abundant AI; native harnesses | 46% | 42.5% | 45% | 31.9% |
Agents & tool use
| Benchmark | Grok 4.7 | GPT-5.6 Sol | Claude Fable 5 (max, with fallback) | Grok 4.6 |
|---|---|---|---|---|
| Legal Agent Benchmark (Harvey)xhigh; Vals AI held-out 120-task run, Valkyrie harness | 19.6% | 2.5% | 11.3% | 15.8% |
| GDPvalxhigh; per launch post | 1695 Elo | n/a | n/a | 1605 |
| AA Briefcase v1.1xhigh; multi-hour office work; per launch post | 1657 points | 1487 | n/a | 1546 |
| EEBenchxhigh; Grok Build harness; Atopile; the launch post table prints 64.0% | 66% | 39.4% | 54.2% | n/a |
Other
| Benchmark | Grok 4.7 | GPT-5.6 Sol | Claude Fable 5 (max, with fallback) | Grok 4.6 |
|---|---|---|---|---|
| CADGenBenchhigh; generation split; Mecado | 44.4% | n/a | n/a | 40.9% |
| CyberGym (cybersecurity)high; mean reproduced; unrestricted setting, Grok Build harness | 80.3% | 83.6% | n/a | 79.7% |
| CVE-Bench (cybersecurity)xhigh reward; 37.7% at high; unrestricted setting | 36.6% | n/a | n/a | 39.8% |
| CathedralBench (cybersecurity)xhigh; hard subset accuracy; unrestricted setting; independent third party | 29% | n/a | n/a | 25% |
Knowledge & science
| Benchmark | Grok 4.7 | GPT-5.6 Sol | Claude Fable 5 (max, with fallback) | Grok 4.6 |
|---|---|---|---|---|
| HealthBench Professionalxhigh; Grok 4.6 used as grader; safety-filtered tasks scored zero | 56.7% | 60.5% | n/a | n/a |
| LatchBio Capabilities v1.0xhigh; mean of 11 agentic biology benchmarks | 44.5% | n/a | n/a | 43.3% |
Safety & honesty
| Benchmark | Grok 4.7 | GPT-5.6 Sol | Claude Fable 5 (max, with fallback) | Grok 4.6 |
|---|---|---|---|---|
| HackerBench v0.3 harmful or dual-use compliance (lower is better)internal CTF-style suite with production safeguards; per launch post | 3.3% | n/a | n/a | n/a |
| BioSecBench-Refusalxhigh; LatchBio; higher is better | 62.4% | n/a | n/a | 45.6% |
| Standard jailbreaks compliance (lower is better)high; production safeguards | 0.01% | n/a | n/a | 0.04% |
| StrongReject compliance (lower is better)high | 2% | n/a | n/a | 3.9% |
| Long-horizon jailbreaks compliance (lower is better)high; multi-turn and Crescendo attacks | 0.65% | n/a | n/a | 1% |
| MASK-Rectified dishonesty (lower is better)high | 0% | n/a | n/a | 1.9% |
| Sycophancy rate (lower is better)high; internal evaluation | 0.03% | n/a | n/a | 0.04% |
| Bio refusal recallhigh; Autointent-Bio | 100% | n/a | n/a | 100% |
| Virology Capabilities Test accuracy (dual-use capability signal)high; lower indicates less misusable capability | 63% | n/a | n/a | 67.4% |
What did safety testing find for Grok 4.7?
Safety level: xAI Frontier Artificial Intelligence Framework (FAIF, 30 June 2026): below the FAIF safety thresholds on dual-use knowledge
- Grok 4.7 scores below the FAIF safety thresholds on dual-use biological knowledge, with no dual-use biological capability increase versus Grok 4.6.
- Cyber: small capability gains over Grok 4.6, concentrated in defence and vulnerability mitigation; CyberGym 80.3% at high versus 79.7%, and third-party evaluators corroborated results on an unrestricted configuration.
- HackerBench v0.3 with safeguards: 3.3% harmful or dual-use compliance per the launch post, with low benign refusal.
- Jailbreak compliance at high effort: 0.01% standard, 2.0% StrongReject and 0.65% long-horizon, against 0.04%, 3.9% and 1.0% for Grok 4.6.
- Self-harm compliance rises to 1.05% from 0.84% for Grok 4.6, and general refusal compliance is 1.10% against 0.93%.
- MASK-Rectified dishonesty is 0.00% (Grok 4.6: 1.90%); child safety compliance is 0.0%.
Independent figures: Artificial Analysis, refreshed daily on our AI model benchmarks page. Full analysis: our Grok 4.7 deep dive.
SpaceXAI Model Timeline
500k tokens context
Specifications
Related Tools
Claude Opus 5.5
Claude Opus 5.5, launched 22 September 2026, is Anthropic's current Opus-line flagship, replacing Opus 5. It performs at roughly the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5, at $4 input / $20 output per million tokens (both 20% down, cache reads cut 60% to $0.20). It leads on Terminal-Bench 4.0, GDPval-AA v2.1 and Humanity's Last Exam, and is the first Opus model to ship with Fable-5.1-class cybersecurity, biology and distillation safeguards.
Claude Sonnet 5.5
Claude Sonnet 5.5, released 28 September 2026, is the second model in Anthropic's Claude 5.5 family and the faster, lower-cost complement to Claude Opus 5.5. It keeps Sonnet 5's $2 input / $10 output per million tokens ($0.20 cache reads) while generating output over 30% faster, and scores 70.6% on Terminal-Bench 4.0, above Opus 5.5's 66.4%. It has a 1M-token context window, 128K max output, adaptive thinking, and is the first Sonnet with cyber and anti-distillation classifiers.
GPT-6.1 Sol
GPT-6.1 Sol, launched at OpenAI DevDay on 29 September 2026, is OpenAI's mid-tier workhorse model. OpenAI says it delivers nearly the same intelligence as GPT-6 Astra at one-fifth of Astra's token price: $2 input / $10 output per million tokens with cached input cut to $0.10. It scores 75.2% on DeepSWE v1.1 (high effort) and 71.4% on OSWorld 2.0, has a 1.05M-token context window and 128K max output, and is available in the API plus ChatGPT Work and Codex.
