SpaceXAI has launched Grok 4.6, saying its model matches leading models from OpenAI and Anthropic while costing less and improving its ability to handle complex, long-running tasks.
The company introduced Grok 4.6 on August 12, positioning it as an upgrade focused on AI agents that can work through complicated jobs over many steps. It said the model is designed for research, coding, software development, analysis and creating interactive applications.
According to SpaceXAI, Grok 4.6 reached frontier-level results across several agentic coding and knowledge-work benchmarks. Artificial Analysis places the model close to OpenAI’s GPT-5.6 Sol on its Intelligence Index, although early comparisons do not establish that it consistently beats leading systems. MarketWatch reported that Grok 4.6 scored five points higher than Grok 4.5 on the index.
Cost is another part of SpaceXAI’s pitch. The model is priced at about $2 per million input tokens and $6 per million output tokens, making it considerably cheaper than some competing frontier models. That could make it attractive to developers running AI agents that consume large amounts of compute.
Grok 4.6 is currently available through Cursor and Grok Build. Its launch also strengthens the relationship between SpaceXAI and Cursor, around software development and agentic coding. Early users on Cursor have reported that the model performs strongly on long coding tasks, although some say it remains below the best models in certain areas.
SpaceXAI said Grok 4.6 was trained using a longer process than Grok 4.5, with synthetic training data covering reasoning, science, technology, software engineering and knowledge work. Grok 4.5 was used to generate additional training examples, while automated checks removed lower-quality data.
The model was then post-trained with reinforcement learning on coding, knowledge work, web development, computer-aided design and software optimisation tasks. SpaceXAI said Grok 4.6 can research a subject, plan an application, build its features, test the result and improve it through repeated feedback.
The company has also expanded safety testing before and after launch. It said these safeguards are intended to reduce risks while allowing the model to work on tasks such as software vulnerability fixes and engineering research.
The release comes as AI companies increasingly compete on agentic capabilities, not simply chatbot performance. For SpaceXAI, Grok 4.6 is a test of whether lower prices and stronger autonomous performance can turn its model into a serious alternative to established leaders.