Not established from the available sources.
XAI · MODEL RELEASE TRACKER
Grok 4.7
Grok 4.7 is a version of xAI's Grok model reported as politically right-leaning and used as the judge model in a Nous Research evaluation of propaganda in Chinese models.CURRENT SNAPSHOT4/5 DIMENSIONS WITH DATA
The dimensions that change the decision.
RELEASE TIMELINE
Published, source-backed release events only.
Artemis Security launches Orion-1, an AI to investigate and respond to corporate cyberattacks
Artemis Security has launched Orion-1, a purpose-built AI model that ingests logs from endpoints, networks, cloud services and SaaS, reconstructs attacks, assigns confidence scores, and can recommend or initiate responses; the company says Orion-1 outperformed general-purpose models like Claude Opus 5.5 and GPT-6 Sol on its internal Decision-Grade Readiness (DGR) benchmark. Orion-1 is in private preview; Artemis says no customer data was used for training and it is unclear whether Orion-1 was built from scratch or fine-tuned from another foundation model.
SpaceXAI releases Grok 4.7 — faster, cheaper model for coding and knowledge work
SpaceXAI released Grok 4.7, a new base-model upgrade positioned as its most capable model for coding and knowledge work; it reportedly uses a larger base model, longer RL training focused on long-running tasks, and an entirely new safeguard stack. The company says Grok 4.7 matches the price and speed of Grok 4.6 while improving verification, longer-context handling, document/presentation generation, benchmark performance (CursorBench 4.0, GDPval, AA Briefcase), and safety metrics (LatchBio biosafety 62.4%, HackerBench v0.3 allowing 3.3% risky prompts); it is available in Cursor, Grok Build, the Grok API, third-party harnesses and cloud platforms, priced from $2 per million input tokens and $6 per million output tokens, with an optional faster (2x) variant at double the price.
VERIFIABLE FACTS
Every value stays attached to a source and date.
Announced and made available by xAI on 2026-09-21; described as "available today" in the announcement.
Positioned as xAI's most capable model for coding and knowledge work.
Built on a new, larger base model compared to Grok 4.6 and trained with a longer reinforcement‑learning run on a harder mix of tasks weighted toward problems that take many hours to complete.
During the experiment, each agent (including Grok 4.7) was allowed up to 40 million tokens and up to twelve simulated weeks.
xAI reports Grok 4.7 is at the frontier in price-performance on CursorBench 4.0, a benchmark that stresses longer-running coding tasks.
xAI reports Grok 4.7 improves on Grok 4.6 on GDPval and AA Briefcase (benchmarks for professional tasks) and performs comparably to other frontier models.
An independent report gives Grok 4.7 a combined score of 46 on the Artificial Analysis Intelligence Index (v4.3.2), placing it mid-pack versus competitors (Claude Fable 5.1 and GPT-6 scored 53).
Reported Terminal-Bench 4.0 agentic-coding score of 26% (lower than certain competitor models reported in the same article).
Researchers queried Grok 4.7 using its API to obtain answers for the study.
xAI positions Grok 4.7 as built for coding, long-running agents, and professional knowledge work (e.g., legal, medical, financial workflows).
The model supports configurable reasoning effort at four levels: low, medium, high, and xhigh.
According to SpaceXAI, Grok 4.7 uses a larger base model than Grok 4.6 and was trained with a longer reinforcement-learning run on a harder mix of tasks weighted toward problems that take many hours to complete.
Reported as politically right-leaning.
xAI reports Grok 4.7 was trained to natively understand the Grok Bot harness, which the company credits for improvements in conversational tasks and general knowledge work.
xAI reports gains across published evaluations, including software engineering benchmarks (CursorBench, DeepSWE), multi-hour terminal and office work (Terminal-Bench, AA Briefcase), and domain benchmarks such as EEBench for electrical engineering and evaluations for legal work.
According to SpaceXAI, Grok 4.7 is trained to work longer on more difficult tasks and "knows when to check its own work."
Grok 4.7 exposes a reasoning effort setting that controls how much deliberation happens before each response.
In experiments reported on Amazon Bedrock, an agent using Grok 4.7 answered 67 of 72 runs correctly across three search-oriented tasks; at the 'xhigh' reasoning effort it answered 18/18 correctly, and at 'low' it answered 15/18 while using about one-eighth of the output tokens.
In the experiment, Grok 4.7 was tested with web browsing capabilities and the ability to send emails autonomously.
Grok 4.7 was one of six AI models evaluated by Vals AI and researcher Matilda Glynn-Henley in an experiment that tasked agents with securing a £350 train refund.
In a study asking models to name one candidate for upcoming races, Grok 4.7 recommended the Republican candidate in 92% of its answers.
Across nearly 500 races in the study, Grok 4.7 and GPT-6.1 Sol agreed on the recommended candidate less than 10% of the time.
Grok 4.7 lost 12 points in Artificial Analysis's overhauled Intelligence Index (version 5).
Reported as the best of 13 frontier models; causal repair = +51.3 points on 62 calibrated systematic failures. (The paper notes the top eight models are statistically indistinguishable.)
Evaluated as a large language model (LLM) agent in multi-agent economic experiments.
Included in experiments on multi-resource trade, bilateral bargaining, and double auctions.
In bargaining experiments, providing exact information led Grok 4.7 to produce less balanced agreements.
In the study's double-auction experiments, full information disclosure inflated price demands and lowered Grok 4.7's welfare from 97% to 62%.
Served as the judge model in Nous Research's evaluation of propaganda in Chinese models, checking whether responses mention, omit, or contradict facts.
WHAT CHANGED