SpaceXAI unveils Grok 4.6 for autonomous AI agents.
The SpaceXAI team unveiled Grok 4.6, a new flagship model focused on long-term agent tasks, programming, and creating interactive applications.
The new version is built on Grok 4.5, released in early July.
According to the statement, developers emphasized not just individual responses but the ability to work long and consistently on a single task. Grok 4.6 can explore unfamiliar subject areas, design an application, implement its core functions, test the result, and then refine it multiple times based on feedback. SpaceXAI also notes that over long trajectories, the model more frequently self-checks its work before moving to the next stage.
Grok 4.6 scored 61 points in the Artificial Analysis Intelligence Index, matching GPT-5.6 Sol in this composite index based on nine benchmarks. Both models are second only to Fable 5 Max.
Source: SpaceXAI.
SpaceXAI extended the additional training phase of the new version compared to Grok 4.5. The process used data selected by the model on reasoning and complex technical concepts.
What’s New in Grok 4.6
One of the main focuses is creating applications from a general idea. Grok 4.6 can independently research requirements, define the project structure, implement interactions, and continue to improve the result.
In visual and interactive projects, SpaceXAI claims a stronger initial result compared to Grok 4.5: the neural network can immediately set the structure and visual language of the application, then proceed to iterations.
Grok 4.6 also received broader training for agent work, especially in scenarios where AI not only responds to requests but independently performs a sequence of actions.
The new model is already available through API, Grok Build, Cursor, and several partner platforms. The base price starts at $2 per million input tokens and $6 per million output tokens. The accelerated version costs twice as much.
Earlier, on August 12, the SpaceXAI team introduced Grok Bot, an AI agent system for performing complex work tasks. The tool operates in a messenger format.
Follow ForkLog on social media
Found a mistake in the text? Select it and press CTRL+ENTER