Grok 4

Performance is soaring! Grok 4.6 is officially released, comparable in strength to GPT-5.6 Sol. Recently, SpaceX AI officially launched its newly upgraded flagship artificial intelligence model Grok4.6. This model focuses on processing complex, long-range Agent tasks, and has demonstrated strong performance comparable to or even surpassing the OpenAI flagship model GPT-5.6Sol in multiple core benchmark tests. Looking back at its technological evolution, SpaceX AI teamed up with Cursor in July to launch Grok 4.5, which is trained with tens of thousands of NVIDIA GB300 GPU computing power. On this basis, Grok4.6 has achieved a further leap. Its training process focuses on comprehensive optimization for long-range complex tasks, using a large amount of model-generated data covering reasoning capabilities, advanced technical concepts, and high-quality engineering code. At the same time, the model has also received extensive Agent reinforcement learning training, covering knowledge work, general programming and multiple specialized domain environments, and can output better initial effects when building visual and interactive applications. In the comprehensive index covering nine popular AI benchmark tests by the authoritative evaluation agency Artificial Analysis, the comprehensive score of Grok4.6 has equaled that of GPT-5.6Sol, and is currently only behind Claude Opus5 and Fable5Max. Specifically, in the GDPval-AA v2 test that evaluates logic and knowledge levels, Grok4.6 achieved an Elo score of 1753; in the AA-Briefcase, a private benchmark test that measures long-term knowledge work agent tasks, its Elo score also reached 1577 points. Both key performances are second only to Claude Opus5. While bringing significant performance upgrades, Grok4.6 continues to maintain a highly competitive operating speed and price system. The model can run at a speed of 80TPS. The input and output prices of the basic version are US$2 and US$6 per million Tokens respectively, while the high-speed variant is US$4 and US$12 per million Tokens. Currently, users of Cursor and Grok Build can already experience it directly, and developers can also access it conveniently through the API. via AI News (author: AI Base)