SpaceXAI Launches Grok 4.6 With Focus On Agentic AI: Here's What's New & Who Gets It

SpaceXAI, the AI company formerly known as xAI, has released Grok 4.6, a new flagship model built around long-running agentic tasks and more ambitious interactive and visual work. The model became available the same day in the Cursor code editor and the company's own Grok Build tool, as well as through its API, with pricing starting at $2 per million input tokens and $6 per million output tokens, according to the company's announcement. Grok 4.6 builds on the previous Grok 4.5 model and extends the company's push into agentic coding and knowledge work.

What is new in Grok 4.6?

xAI describes Grok 4.6 as a model that stays with complex tasks across many steps, whether researching a topic, working across an entire codebase, or turning a broad product idea into a working first version of an application. On longer task sequences, the company says it observed more self-testing and verification behaviour, with the model checking its own work before moving on, along with stronger first attempts on visual and interactive projects compared with Grok 4.5.

The capability gains come from a longer supplemental training run than Grok 4.5 received. xAI used curated, model-generated data for reasoning and technical concepts, alongside an improved optimiser and supervised fine-tuning trajectories regenerated by Grok 4.5 across multiple reasoning efforts, agent harnesses and domains, including STEM, software engineering and general knowledge work, with model-based checks used to filter out problematic traces. The model was then trained on a range of agentic reinforcement learning tasks, including domain-specific environments for kernel optimisation, web development and computer-aided design.

On benchmarks, xAI reports that Grok 4.6 matches OpenAI's GPT-5.6 Sol on the Artificial Analysis Intelligence Index, a composite of nine benchmarks, with both models scoring 61, behind Anthropic's Fable 5 Max at 62 and ahead of Grok 4.5 High at 56.

On individual tests, Grok 4.6 scored 65.9 percent on the DeepSWE v1.1 software engineering benchmark, up from 54 percent for Grok 4.5, though still behind GPT-5.6 Sol Max at 73 percent. On Terminal-Bench v3.0, it reached 26 percent, a sizeable jump from Grok 4.5's 15.7 percent, but behind the roughly 34 percent posted by GPT-5.6 Sol Max and Fable 5 Max.

The model led on the GDPVal-AA v2 knowledge work evaluation with a score of 1753, and also posted gains on the CursorBench and FrontierCode benchmarks. These figures are xAI's own, and comparative scores for rival models are drawn from self-reported or publicly available results, with no independent verification yet available.

Grok 4.6: Rollout timeline and availability

Grok 4.6 is available from launch in the Cursor code editor and in Grok Build, xAI's own application-building tool, as well as through the company's API. The model is also accessible through third-party partners including OpenRouter, Vercel and Cloudflare. To encourage developer adoption, xAI is offering double the included usage inside Grok Build and Cursor for the first week following launch. The release positions Grok 4.6 as the first flagship model to carry the SpaceXAI name from the ground up, with its pricing, partner distribution and agentic focus reflecting the division's shift toward positioning Grok as a developer and enterprise platform rather than a feature built primarily for the X social network.

Published on: Thursday, August 13, 2026, 11:40 AM IST