SpaceXAI released Grok 4.6, a large language model that scored 61 on the Artificial Analysis Intelligence Index, matching OpenAI's GPT-5.6 Sol and one point behind Claude Fable 5.
The model was trained with AI-generated datasets and reinforcement learning to improve reasoning, and shows particular strength in software prototyping and long-horizon project work.
It is priced at $2 per million input tokens and $6 per million output tokens, and is available through Cursor and Grok Build.
What happened
SpaceXAI released Grok 4.6, a large language model built with extended training on AI-generated datasets and reinforcement learning. The model scored 61 on the Artificial Analysis Intelligence Index, matching OpenAI's GPT-5.6 Sol and one point behind Anthropic's Claude Fable 5.
Why it matters
SpaceXAI (formerly xAI, acquired by SpaceX and publicly listed in June) is rapidly iterating on its flagship LLM—Grok 4.6 arrived just a month after its predecessor. The new model shows particular strength in software prototyping, visual asset creation, and self-correction on long-horizon projects, areas where it outperformed Claude Fable 5 on benchmarks like AA-Briefcase.
What to watch
Grok 4.6 is priced at $2 per million input tokens and $6 per million output tokens (a faster version costs twice as much), and is available through Cursor (the coding platform SpaceXAI acquired for $60 billion in June) and Grok Build.
SpaceXAI unveiled Grok 4.6, its latest large language model, in a release that underscores the company's accelerating development timeline. The model arrived roughly one month after SpaceXAI released its previous flagship LLM, with the company attributing the advancement to extended training using AI-generated datasets specifically designed to enhance reasoning. The training process involved multiple stages: an initial extended run with high-quality engineering data, followed by supervised fine-tuning (SFT) optimized for science and programming tasks, and finally reinforcement learning to refine outputs.
SpaceXAI evaluated Grok 4.6 on the Artificial Analysis Intelligence Index, a composite benchmark combining nine popular AI benchmarks across science, coding, and financial services. The model scored 61, placing it on par with OpenAI's GPT-5.6 Sol but one point behind Anthropic's Claude Fable 5. When tested across nine additional benchmarks spanning more than a half-dozen industries, Grok 4.5 (the predecessor) outperformed Claude Fable 5 in three evaluations. The company emphasized that Grok 4.6 excels at generating software prototypes from high-level descriptions, creating visual assets such as interfaces, and checking its own work for errors on long-horizon projects.
The model is available in two tiers: a standard version priced at $2 per million input tokens and $6 per million output tokens, and a faster edition at double the cost. Access is provided through Cursor, the coding platform SpaceXAI acquired for $60 billion in June, and through Grok Build, an internally developed programming tool. This distribution strategy embeds the model directly into SpaceXAI's developer ecosystem, leveraging its recent acquisition spree to expand reach and usage.
SpaceXAI, rebranded from xAI following its acquisition by SpaceX and public listing in June, is moving rapidly to establish itself as a competitive AI player. The release of Grok 4.6 just one month after its predecessor signals an aggressive development cadence. The company invested heavily in extended training with AI-generated datasets designed to improve reasoning—a core capability gap in earlier models—and followed this with supervised fine-tuning optimized on tasks like science and programming, then reinforcement learning to further refine outputs.
The model's positioning against Claude Fable 5 and GPT-5.6 Sol is deliberate. While Grok 4.6 matches GPT-5.6 Sol on the Artificial Analysis Intelligence Index (both at 61), it trails Claude Fable 5 by one point. However, the company highlights wins in specific domains: Grok 4.5 outperformed Claude Fable 5 on three of nine evaluated benchmarks, and Grok 4.6 is characterized as particularly strong at software prototyping, visual asset generation, and error-checking on extended projects—capabilities that could appeal to developers and engineering teams. The availability through Cursor (acquired for $60 billion) and Grok Build positions the model within SpaceXAI's own developer infrastructure.
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytime
Ask AI anything about this article. Q&As are published on this page for other readers too.
Google launched the Pixel 11, Pixel 11 Pro, Pixel 11 Pro XL, Pixel 11 Pro Fold, and Pixel Watch 5
Antrique Holdings, through its Brunei subsidiaries, unveiled an AI Innovation Platform on August 13, 2026, mar…
Naver has reportedly invested in Anthropic, a US artificial intelligence developer
SpaceX's xAI released Grok Bot, an AI agent in early beta that signs into your tools and completes work autono…
Anthropic's Claude AI, working on the unsolved mathematics problem known as the Riemann hypothesis, initially…
DeepSeek released V4 Pro 0813, its latest Pro model, available through OpenRouter via API
The AI news that matters, in one minute each morning.
Sign up free