SpaceXAI, the artificial intelligence company formerly known as xAI, released Grok 4.6 on August 12, 2026, a new flagship model built for long-running agents and more ambitious interactive and visual work. The model is available the same day in the Cursor code editor and the company’s own Grok Build tool, as well as through its API, at a starting price of $2 per million input tokens and $6 per million output tokens, according to the company’s announcement.
Grok 4.6 builds on Grok 4.5 and extends the company’s push into agentic coding and knowledge work.
What Grok 4.6 Does
The company describes Grok 4.6 as a model that “stays with complex tasks across many steps,” whether researching a topic, working across a codebase, or turning a broad product idea into a working first version of an application. On longer trajectories, the company says it observed more self-testing and verification, with the model checking its own work before moving on, and stronger first passes on visual and interactive projects than it saw with Grok 4.5.
The capability gains rest on a longer supplemental training run than Grok 4.5 received. SpaceXAI used curated model-generated data for reasoning and technical concepts, an improved optimizer, and SFT trajectories regenerated by Grok 4.5 across reasoning efforts, agent harnesses, and domains including STEM, software engineering, and knowledge work, with problematic traces filtered out by model-based checks. The model was then trained on a range of agentic reinforcement-learning tasks, including domain-specific environments for kernel optimization, web development, and computer-aided design.
The Benchmark Race
SpaceXAI 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. The company’s own evaluation table shows a more mixed picture across individual tests.
On the DeepSWE v1.1 software-engineering benchmark, Grok 4.6 scored 65.9%, up from Grok 4.5’s 54% but trailing GPT-5.6 Sol Max at 73%. On Terminal-Bench v3.0 it reached 26%, a large jump from Grok 4.5’s 15.7% but well behind the roughly 34% posted by GPT-5.6 Sol Max and Fable 5 Max. It led the field on GDPVal-AA v2, a knowledge-work evaluation, at 1753, and posted gains on CursorBench and FrontierCode. The figures are the company’s own, with third-party model scores drawn from self-reported or publicly available results, and no independent evaluation has yet confirmed them.
The launch lands in a crowded part of the market. Anthropic has been pushing its Opus line toward frontier intelligence at aggressive prices, and OpenAI’s GPT-5.6 family is the incumbent target for agentic coding. SpaceXAI’s strategy is to compete on price and distribution as much as raw capability: at $2 per million input tokens, Grok 4.6 undercuts many frontier rivals, and the company is offering double the included usage inside Grok Build and Cursor for the first week to pull developers onto the model. Beyond Cursor and Grok Build, it is also available through partners including OpenRouter, Vercel, and Cloudflare.
A New Company Behind the Model
Grok 4.6 is the clearest product statement yet from a business that has been entirely restructured over the past year.
Grok 4.6 is the first flagship model to carry the SpaceXAI name from the ground up, and its pricing, partner distribution, and agentic focus show the division positioning Grok as a developer and enterprise platform rather than a feature of the X social network.
