SpaceXAI has released Grok 4.6, a new version of its AI model that targets long-running agentic tasks, software development and interactive applications.
Grok 4.6 is available through Cursor, Grok Build and xAI’s API, as well as platforms including OpenRouter, Vercel and Cloudflare. Pricing starts at $2 per million input tokens and $6 per million output tokens. A faster variant is priced at twice those rates.
“Grok 4.6 builds on Grok 4.5 with a particular focus on long-running agents and more ambitious interactive and visual work,” SpaceXAI said in its announcement.
The company said the model can sustain tasks across multiple steps, including researching topics, analysing information, working across codebases and building applications.
On the Artificial Analysis Intelligence Index, Grok 4.6 scored 61, matching GPT-5.6 Sol and trailing Fable 5’s score of 62. The index combines results from nine benchmarks.
Grok 4.6 also scored 69.9% on CursorBench v3.2, compared with 67.2% for GPT-5.6 Sol and 70.5% for Fable 5. On DeepSWE v1.1, it scored 65.9%, while GPT-5.6 Sol scored 73% and Fable 5 scored 70%.
The model performed better than Grok 4.5 across the benchmarks published by xAI. Its score on DeepSWE increased to 65.9% from 54%, while its CursorBench score rose from 66.7% to 69.9%.
SpaceXAI said Grok 4.6 underwent a longer supplemental training run than Grok 4.5 using model-generated data covering reasoning and technical concepts, engineering data, and an updated training recipe.
The company also used Grok 4.5 to generate supervised fine-tuning trajectories across different reasoning efforts, agent environments and areas including STEM, software engineering and knowledge work.
Grok 4.6 was then trained with reinforcement learning on tasks covering coding and knowledge work, along with environments for kernel optimisation, web development and computer-aided design.
SpaceXAI said the model is particularly suited to turning broad product ideas into working applications. It can research unfamiliar areas, structure an application, implement interactions and refine the output through multiple rounds of feedback.
“On longer trajectories, we also started to see more self-testing and verification, with the model checking its own work before moving on,” the company added.
The model is also being positioned for visual and interactive development. xAI said Grok 4.6 can establish the structure and visual language of an application in a single pass before iterating on the result.
The release comes as AI companies increasingly position their models as agents capable of completing multi-step software and knowledge-work tasks rather than responding to individual prompts.
SpaceX CEO Elon Musk also said that the company is planning to launch Grok 4.7 soon. “Grok 4.7 is significantly better than 4.6 and should be ready in three to four weeks. Initial training is complete, and now we’re adding a massive amount of SpaceX company data in supplemental training. This will be something special,” he wrote in a post on X.
Your reaction
Discussion
What's your take on this story?
No comments yet — be the first to share your view.