Published: 13 August 2026
xAI, which rebranded to SpaceXAI following its acquisition by SpaceX, released Grok 4.6 on 12 August 2026, a new flagship large language model built for long-running agentic work and more ambitious interactive and visual tasks. According to third-party benchmark site Artificial Analysis, Grok 4.6 scores 61 on the Artificial Analysis Intelligence Index, matching OpenAI’s GPT-5.6 Sol and trailing Anthropic’s Claude Fable 5 by one point.
What can Grok 4.6 do?
xAI positions Grok 4.6 for tasks that stay open across many steps, such as researching a topic, analysing information, working across a codebase, or turning an idea into a finished application or work artifact. The company says the model is particularly strong at generating software prototypes from high-level descriptions and at producing visual assets such as interfaces, and that it is more likely than its predecessor to check its own work for errors on long-horizon projects.
Grok 4.6 rolled out roughly a month after xAI’s previous flagship model, Grok 4.5, a faster release cadence than the company’s earlier model generations.
How was Grok 4.6 trained?
xAI says one of the main changes behind Grok 4.6 is that engineers spent more time on the training run itself, using an AI-generated dataset designed to improve the model’s reasoning and giving it access to what the company describes as high-quality engineering data. This description comes from xAI and has not been independently verified.
The initial training run was followed by two further stages: supervised fine-tuning (SFT), which refines a model’s output format using sample prompts and pre-packaged answers, and reinforcement learning. xAI used the outgoing Grok 4.5 model to help optimise Grok 4.6’s SFT phase, with that optimisation focused specifically on science and programming tasks.
Grok 4.6 benchmarks and technical specs
- On the Artificial Analysis Intelligence Index, a third-party composite of nine benchmarks spanning science, coding and financial services tasks, Grok 4.6 scores 61, level with OpenAI’s GPT-5.6 Sol and one point behind Anthropic’s Claude Fable 5. Only Claude Fable 5 and Claude Opus 5 score higher among named competing models.
- On AA-Briefcase, Artificial Analysis’s private benchmark of long-horizon agentic knowledge-work projects, Grok 4.6 debuts at an Elo of 1577, behind Claude Opus 5.
Grok 4.6 has a 500,000-token context window and a knowledge cutoff of 1 February 2026. These benchmark results are reported by Artificial Analysis rather than xAI itself, and xAI has not published its own full evaluation harness for the comparisons it cites in its announcement.
How does Grok 4.6 compare to GPT-5.6 Sol and Claude Fable 5?
On the Artificial Analysis Intelligence Index, Grok 4.6 sits level with GPT-5.6 Sol and one point behind Claude Fable 5, placing it among the top handful of frontier models on that index alongside Claude Opus 5. xAI also says Grok 4.6 outperforms Claude Fable 5 on three of nine additional benchmarks it evaluated, including tasks that span more than half a dozen industries, though xAI has not published the full harness for those additional comparisons.
Grok 4.6 availability and pricing
Grok 4.6 is available the same day through Cursor, the coding platform that SpaceXAI acquired for 60 billion dollars in June 2026, through xAI’s own Grok Build programming tool, and through xAI’s API. The standard version is priced at 2 dollars per million input tokens and 6 dollars per million output tokens; a faster edition is available at twice that price.
For the full details please visit the official xAI announcement of Grok 4.6.
