SpaceXAI unveils Grok 4.6, claims frontier-level performance at lower cost
On August 12, SpaceXAI unveiled its latest AI model, Grok 4.6, claiming it matches the performance of frontier models from Anthropic and OpenAI at a lower cost. The new model surpasses Grok 4.5, delivering superior performance while costing less than Fable 5 and GPT-5.6. Grok 4.6 focuses on long-running agents and complex, interactive visual work, excelling at multi-step tasks like researching topics, analyzing information, working with codebases, and transforming ideas into polished applications or work artefacts.
Grok 4.6 achieved frontier intelligence levels across several agentic coding and knowledge work benchmarks, matching GPT-5.6 Sol's performance on the Artificial Intelligence Analysis Index. Available in Cursor and Grok Build, the model was trained using an advanced process emphasizing reasoning, STEM, software engineering, and knowledge work.
SpaceXAI improved Grok 4.5 training methods and optimization, then post-trained Grok 4.6 through reinforcement learning (RL) on coding, knowledge work, web development, computer-aided design, and software optimization.
One of Grok 4.6's key improvements is its ability to convert complex ideas into functional projects, researching topics, planning applications, building features, and refining results through multiple feedback rounds. The model also checks and tests its work, producing better first versions of visual and interactive applications than Grok 4.5.
SpaceXAI has expanded Grok 4.6's safety testing, incorporating broader pre-launch evaluations, post-launch, and third-party testing to balance usefulness and security, including tasks like fixing software vulnerabilities and supporting engineering and AI research.
Written by urgent.news from The Indian Express's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
