Industry / Models / Grok / x.ai
Grok
Coding and knowledge-work models for tool-driven engineering and operational assistants.
- Model snapshot
- Grok 4.7
- Deployment
- Grok API and supported agent or coding environments
- Context & outputs
- Endpoint-specific; see the model reference
- Access & license
- Proprietary hosted model
Coding and knowledge work
Grok 4.7 is positioned for longer-running coding and professional tasks, including self-checking and context management. Its release table reports coding, terminal, and office-work evaluations with effort-level notes. Grok 4.7 release.
Tools and grounding
The developer platform documents structured outputs, function calling, search, code execution, and remote MCP tools. Those are configured capabilities of a workflow; they do not mean every response has automatically searched current sources. Grok developer reference.
Our engineering approach
We consider Grok for engineering agents and assistants that act through explicit tools. We test the complete path from retrieval to action, preserve source attribution, and isolate tools that can change external state. For coding, we measure successful changes and regressions. For knowledge work, we measure evidence quality and artifact correctness. We compare elapsed time and total task cost at the selected effort level rather than assuming a faster token stream yields a better workflow.
Published evidence
Benchmark snapshot.
Provider-reported Grok 4.7 results from September 21, 2026. The table uses xHigh effort except the starred DeepSWE result, which uses high. AA Briefcase is an Elo score, not a percentage.
| Parameter / benchmark | Result | Evaluation setting | Source |
|---|---|---|---|
| Software engineeringCursorBench 4.0 | 46.3% | Grok 4.7 xHigh | Published result |
| Software engineeringDeepSWE v1.1 | 71.0% | High effort, per provider footnote | Published result |
| Terminal agentsTerminal-Bench 4.0 | 37.6% | Grok 4.7 xHigh | Published result |
| Knowledge workAA Briefcase v1.1 | 1,657 Elo | Grok 4.7 xHigh | Published result |
These are published results, not Glowing Spoon measurements. Benchmark percentages, F1 scores, and Elo ratings use different scales. Production evaluation also includes latency, cost, reliability, and human review effort on your workload.