Grok 4.7
Hosted frontier model for coding and knowledge work; the developer describes a new base model and extended reinforcement-learning run.
Score calculation
(4 + 4 + 2 + 3) ÷ 4 = 3.3
4 of six dimensions rated. Unrated dimensions and BDS status are excluded.
Full scoring method ↗Setup details
Release notes are Grok 4.7-specific; privacy summaries concern consumer Grok and X, and should not be assumed to govern every API or partner deployment.
xAI announced Grok 4.7 on September 21, 2026 and lists it as the current flagship in its model catalog. Checked 2026-10-23.
Energy
↑xAI's release describes speed and price improvements but gives no Grok 4.7 training energy, inference energy, water, or carbon figures. Throughput or token-cost claims are not environmental measurements.
Reasoning & sources
The Grok 4.7 release reports speed and price improvements, neither of which measures electricity, emissions, or water per task. A faster model may change total demand as well as per-task use; do not infer either direction without workload-level measurements.
No model-specific compute or environmental report found.
Training rights
↑xAI says Grok pretraining primarily uses publicly available information, with quality filters and efforts to minimize personal data. Its 4.7 release identifies reinforcement learning but does not publish a dataset-level rights inventory.
Reasoning & sources
xAI says Grok pretraining mainly uses public information with filters and describes reinforcement-learning data, but gives no source-level corpus or rights audit for 4.7. These are process descriptions, not proof that particular training sources were licensed; users needing traceability should request provenance and opt-out details.
Publicly available training information and filtering are stated, but rights and corpus detail are not listed.
Source-level dataset, licenses, consent and rights-resolution process are not disclosed in the checked sources.
Privacy
↑Grok consumer content may be used for training unless users opt out; Private Chat is excluded from training and is deleted within 30 days. X separately says public posts and Grok interactions may be shared with xAI, with controls and exceptions. API terms are separate.
Reasoning & sources
The cited xAI and X notices say consumer Grok data may train models unless users opt out, while Private Chat has separate no-training and deletion terms. Those controls are specific to consumer surfaces and do not automatically govern API or third-party hosting; check the service-specific policy.
Consumer training opt-out and Private Chat controls are documented, with surface/geography exceptions.
Controls differ by logged-in status, geography, product surface, and feedback submission.
Openness
↑Grok 4.7 is a proprietary hosted/API model. The release identifies model changes at a high level but publishes no weights or full training pipeline.
Reasoning & sources
The 4.7 release and developer catalog provide hosted/API access but no downloadable weights or reproducible training pipeline. This supports a low openness judgment; provider access should not be described as open access.
Proprietary hosted/API release without public weights.
No public weights or full training recipes found.
Worker treatment
↑X says Grok training data is curated by human AI Tutors and the release describes reinforcement learning. The checked sources do not state tutor/annotator employment conditions, pay, location, contractors, or grievance mechanisms.
Reasoning & sources
xAI says human AI Tutors curate Grok training data and the release describes reinforcement learning, but neither source gives tutor compensation, location, contractor status, or conditions. Human involvement is documented, but this does not establish worker protections; request terms before drawing a labor conclusion.
No worker-condition disclosure found.
Government & military
↑xAI’s government page says its products were made available to US government customers in July 2025 and names Grok among the family used by government and national-security customers; a current government page describes SpaceXSI use of the same models in classified and unclassified contexts for intelligence analysis, operational planning, and defense logistics. These disclosures do not establish Grok 4.7 as the exact deployed checkpoint.
Reasoning & sources
xAI’s government pages explicitly describe its model family in government, classified, intelligence-analysis, planning, and defense-logistics settings, without naming Grok 4.7 as the exact checkpoint. This merits a serious company/family-level concern, while procurement screening should still ask which version and deployment are in scope.
xAI directly describes government and defense/intelligence deployments of its model family, though the exact current checkpoint is not specified.
The checked pages do not provide a model-by-model contract or deployment register, nor do they identify Grok 4.7 in the described government workflows.
BDS status
↑We did not find xAI among the named targets in the two official BNC campaign lists. This is a limited list check, not BDS approval.
Reasoning & sources
xAI is not a named target in the checked lists, but BDS uses selective campaigns, so absence cannot establish a clean record. Running downloaded weights locally is a different decision from paying the developer or a cloud host; check the host as well.
This review covers the named targets in the BNC consumer/pressure guide and No Tech campaign, not every BDS campaign, local group, investment or supplier. Lists and ownership can change. A company’s absence from these lists is not an endorsement or proof that it has no relevant relationships.