Elon Musk’s SpaceXAI (formerly xAI) unfurled Grok 4.6, the company’s latest flagship artificial intelligence (AI) model designed specifically for long-running agents, complex software engineering, and knowledge work – but at a steep discount.
The release last week marks a swift iteration for the AI firm, arriving just weeks after Grok 4.5 and following the debut of Grok Bot, a system engineered to deploy AI agents as virtual employees capable of managing complex, multi-step tasks.
According to evaluations by Artificial Analysis, Grok 4.6 achieves a score of 61 on its Intelligence Index. The result places the new model in a tie with OpenAI’s GPT-5.6 Sol Max, surpassing Moonshot AI’s Kimi K3 while trailing Anthropic’s top-ranked Claude Opus 5 and Fable 5 models.
Despite its competitive standing among top-tier models, SpaceXAI is positioning Grok 4.6 with an aggressive pricing structure aimed at enterprise adoption. The model’s application programming interface (API) starts at $2 per million input tokens and $6 per million output tokens — less than half the cost of OpenAI’s standard GPT-5.6 Sol API rates.
On specialized benchmarks, Grok 4.6 demonstrates notable generational gains over Grok 4.5.
On the GDPVal-AA v2 benchmark, Grok 4.6 scored 1,753, outperforming both GPT-5.6 Sol Max (1,728) and Fable 5 Max (1,741). The model reached 69.9% on CursorBench v3.2 and 65.9% on DeepSWE v1.1, showing marked improvement, though trailing GPT-5.6 Sol Max’s 73% lead in software development tasks. And on APEX-Agents, Grok 4.6 posted 57.5%, edging past GPT-5.6 Sol Max (56.7%).
SpaceXAI noted that the model was subjected to expanded supplemental training using synthetic reasoning, technical data, and targeted reinforcement learning across environments like kernel optimization, web development, and computer-aided design. The result is a model designed to maintain state, self-test, and execute multi-step tools over extended periods.
Following its Wednesday release, SpaceXAI moved quickly to integrate Grok 4.6 across major developer platforms. On Friday, GitHub announced that Grok 4.6 is now rolling out directly to GitHub Copilot across eight primary environments, including VS Code, Visual Studio, Xcode, JetBrains IDEs, and the Copilot CLI.
The rollout spans Copilot Pro, Pro+, Max, Business, and Enterprise tiers. On enterprise plans, administrators must manually enable access to the new model. Within Copilot, Grok 4.6 is charged under usage-based billing at provider list prices, offering developers a significantly lower-cost alternative to competing options like Claude Opus and GPT-5.6 Sol.
Beyond GitHub, Grok 4.6 is available immediately via SpaceXAI’s own Grok Build platform—included in the $30 monthly SuperGrok subscription—and Cursor, the AI coding startup recently acquired by SpaceX. Third-party cloud providers, including OpenRouter, Vercel, and Cloudflare, have also added support for the model.
To incentivize early adoption, SpaceXAI is offering double the included usage allowance for Grok 4.6 across Cursor and Grok Build through Aug. 19.
