SpaceXAI has introduced Grok 4.6, the latest version of its AI model, with a focus on long-running agentic tasks, software development and interactive applications.
Grok 4.6 is now available through Cursor, Grok Build and the xAI API, along with platforms such as OpenRouter, Vercel and Cloudflare. Pricing starts at $2 per million input tokens and $6 per million output tokens, while its faster version costs twice as much.
According to SpaceXAI, Grok 4.6 builds on Grok 4.5 with improvements designed to help AI agents handle complex, multi-step tasks. The model can research unfamiliar topics, analyse information, work across large codebases and develop applications through multiple stages.
On the Artificial Analysis Intelligence Index, Grok 4.6 received a score of 61, matching GPT-5.6 Sol and slightly below Fable 5, which scored 62. The index evaluates models across nine different benchmarks.
Grok 4.6 also recorded a 69.9% score on CursorBench v3.2, compared with 67.2% for GPT-5.6 Sol and 70.5% for Fable 5. On DeepSWE v1.1, Grok 4.6 achieved 65.9%, while GPT-5.6 Sol scored 73% and Fable 5 reached 70%.
The new model showed clear progress over Grok 4.5 in xAI’s reported tests. Its DeepSWE score increased from 54% to 65.9%, while its CursorBench performance improved from 66.7% to 69.9%.
SpaceXAI said Grok 4.6 received additional training using model-generated data focused on reasoning, technical concepts, engineering and other areas. Grok 4.5 was also used to create supervised fine-tuning examples covering different reasoning levels, agent environments, STEM and software engineering.
The company then applied reinforcement learning to tasks involving coding and knowledge work, as well as environments focused on kernel optimisation, web development and computer-aided design.
A major focus of Grok 4.6 is turning broad ideas into usable applications. The model can research a subject, plan an application, build its features and improve the result through repeated feedback. SpaceXAI also said the model has shown stronger self-testing and verification during longer tasks.
Grok 4.6 is also aimed at visual and interactive development. xAI says the model can establish an application’s structure and visual style in a single attempt before making further improvements.
The launch reflects a wider shift in the AI industry toward models that can complete extended software and knowledge-work tasks instead of simply answering individual prompts.
Meanwhile, SpaceX CEO Elon Musk has indicated that Grok 4.7 could arrive within three to four weeks. Musk said initial training has been completed and that additional SpaceX company data is being added during supplemental training.
