X.AI Launches Grok 4.6 with Enhanced Agentic Capabilities

X.AI released Grok 4.6 on Monday, expanding its agentic AI lineup. The new version builds directly on Grok 4.5 and places particular emphasis on long‑running agents and more ambitious interactive and visual work. According to the company, Grok 4.6 stays with complex tasks across many steps, whether researching a topic, analyzing information, working across a codebase, or turning an idea into a polished application or work artifact.

The model achieves frontier intelligence across several agentic coding and knowledge work benchmarks. It matches GPT‑5.6 Sol on the Artificial Analysis Intelligence Index, which combines scores from nine separate benchmarks into a single composite figure. This parity indicates that Grok 4.6 performs at a level comparable to the latest GPT‑5.6 Sol variant on those tests.

To encourage early adoption, X.AI is offering double the usual included usage inside Grok Build and Cursor for the first week. Users can therefore try the model immediately without consuming their standard allocation.

Grok 4.6 is available today in Cursor and Grok Build. It is also accessible through the API and via partners such as OpenRouter, Vercel, and Cloudflare. Pricing starts at $2 per million input tokens and $6 per million output tokens, with a fast variant priced at twice those rates.

The company says Grok 4.6 underwent a longer supplemental training run than its predecessor. That run used curated model‑generated data for reasoning and advanced technical concepts, high‑quality engineering data, and an improved optimizer and training recipe. The resulting foundation strengthened the subsequent supervised fine‑tuning and reinforcement learning stages.

During development, X.AI used Grok 4.5 to regenerate supervised fine‑tuning trajectories across reasoning efforts, agent harnesses, and domains including STEM, software engineering, and knowledge work. Problematic traces were filtered out with model‑based checks, producing a checkpoint that shows strong performance and improved behavior.

Grok 4.6 was trained on a wide range of agentic reinforcement‑learning tasks, covering knowledge work, general coding, and domain‑specific environments such as kernel optimization, web development, and computer‑aided design. Testing on projects designed to stretch the model’s range and ability to sustain work over many steps revealed particular strength in turning a broad product idea into a working first version. The model can research unfamiliar domains, structure an application, implement core interactions, and refine the result through several rounds of feedback.

On longer trajectories, Grok 4.6 exhibits more self‑testing and verification, checking its own work before proceeding. The company notes that the model produces stronger first passes on visual and interactive projects than Grok 4.5 typically did. Given a concrete product idea, it can establish structure and visual language for an application in a single pass, making it useful for teams that prefer to start with a substantial prototype and then iterate.

Safeguards have been updated to align with the model’s expanded capabilities. X.AI says its safety stack is designed to maximize utility and security across legitimate use cases, allowing Grok 4.6 to be helpful and safe in areas such as vulnerability patching, accelerating the engineering design cycle, and augmenting AI research. The safeguard evaluation includes the widest‑ever suite of pre‑deployment testing for capabilities and calibration, as well as extensive post‑deployment and third‑party testing.

The launch positions Grok 4.6 as a direct competitor to leading general‑purpose models in the agentic and knowledge‑work space, with measurable benchmark parity and a clear path for developers to begin building with increased usage allowances.