xAI has released Grok 4.6 publicly, claiming the new model reaches frontier scores on several agentic coding and knowledge work benchmarks and matches OpenAI’s GPT-5.6 Sol on a number of the same tests, including the Artificial Analysis Intelligence Index, a composite score drawn from nine benchmarks. The release marks Grok’s return to competition with OpenAI and Anthropic after a period in which xAI’s models had lagged behind.<\/p>\n
xAI says Grok 4.6 achieves frontier intelligence across several agentic coding and knowledge work benchmarks. On the Artificial Analysis Intelligence Index, which aggregates nine benchmarks into a single composite score, the company says Grok 4.6 matches GPT-5.6 Sol rather than surpasses it. Elon Musk, writing on X, went further than the official announcement and described Grok 4.6 as "objectively #1 when considering intelligence, speed [and] cost," a framing that goes beyond matching a competitor on a single composite index.<\/p>\n
xAI’s recent benchmark recovery is tied to its relationship with Cursor, the agentic coding company. Grok 4.5 was the first version of xAI’s model trained in part on real-world usage data from Cursor, and Grok 4.6 reportedly went through an even longer supplemental training run on the same data source than its predecessor. xAI partnered with Cursor and may have acquired the company in the months before SpaceX went public, which gives xAI ongoing access to a stream of coding telemetry that most rival labs do not have.<\/p>\n
The partnership shows up directly in how Grok 4.6 is being distributed. Grok 4.6 is available first in Cursor, alongside Grok Build, xAI’s coding agent. Cursor and xAI have also jointly shipped an always-on AI agent called Grok Bot, released in beta earlier this week.<\/p>\n
Better benchmark numbers do not yet translate into broader market position. Earlier versions of Grok drew widespread criticism after being used to generate large volumes of non-consensual nude images, including images of children. Federal agencies have raised concerns about the model’s safety, and Grok has not become a default option inside the federal government even though Musk spent roughly $400 million supporting the second Trump administration.<\/p>\n
Enterprise adoption is also small. According to Ramp’s AI Index, just 4% of companies that have adopted AI tools are paying for xAI’s option, which is consistent with broader reporting that has framed Grok as a niche product. An earlier favorable perception is a useful base for any model, but the niche Grok currently occupies is tied to controversy rather than to enterprise contracts.<\/p>\n
Three near-term signals will show whether the Cursor-driven training lift is sticking. First, independent third-party benchmark runs on Grok 4.6, because xAI’s own framing emphasizes matching GPT-5.6 Sol rather than beating it. Second, the trajectory of Grok Bot out of beta, since the always-on agent design is the most visible product expression of the Cursor partnership. Third, any movement in the Ramp AI Index share figure, which is currently the cleanest public read on whether enterprise buyers are switching to Grok in volume.<\/p>\n
Grok 4.6 is the latest publicly released model from xAI. xAI says it reaches frontier scores on several agentic coding and knowledge work benchmarks.<\/p>\n
xAI says Grok 4.6 matches GPT-5.6 Sol on a number of benchmarks, including the Artificial Analysis Intelligence Index, a composite score of nine benchmarks. Elon Musk has claimed Grok 4.6 is "objectively #1 when considering intelligence, speed [and] cost."<\/p>\n
Cursor, an agentic coding company that xAI partnered with and possibly acquired, provided real-world coding usage data. Grok 4.5 was the first Grok version trained in part on that data, and Grok 4.6 reportedly had a longer supplemental training run on it. Grok 4.6 is available first in Cursor, alongside Grok Build, and the two companies have jointly shipped Grok Bot in beta.<\/p>\n