SpaceXAI Grok 4

The performance of SpaceXAI Grok 4.6 model has equaled or even surpassed GPT-5.6 Sol SpaceXAI officially released the newly upgraded flagship AI model Grok 4.6 on August 12. This model is purpose-built to handle complex, long-range Agent tasks, and has demonstrated strong performance comparable to or even exceeding the OpenAI flagship model GPT-5.6 Sol in multiple benchmark tests. Previously in July, SpaceXAI teamed up with Cursor to launch the Grok 4.5 model. The model is trained with the computing power of tens of thousands of NVIDIA GB300 GPUs. Its performance is second only to Anthropic's Fable 5 and OpenAI's GPT-5.6 Sol, and it is extremely cost-effective. The Grok 4.6 launched this time has achieved a further leap on this basis. The training process is optimized for long-range complex tasks, using selected models to generate data, covering reasoning capabilities, advanced technical concepts and high-quality engineering code. In addition, the model is trained on extensive agent reinforcement learning, including knowledge work, general programming, and multiple domain-specific environments, giving it better initial results when building visual and interactive applications. According to Artificial Analysis's comprehensive index covering nine popular AI benchmarks, Grok 4.6's performance has equaled that of GPT-5.6 Sol, trailing only Claude Opus 5 and Fable 5 Max. In the GDPval-AA v2 test that evaluates logic and knowledge levels, Grok 4.6 achieved an Elo score of 1753; in the AA-Briefcase, a private benchmark that measures long-term knowledge work Agent tasks, its Elo score also reached 1577 points, both performances second only to Claude Opus 5. While performance has been significantly improved, Grok 4.6 still maintains a highly competitive speed and price system. The model runs at 80 TPS, with input and output prices of $2 and $6 per million Tokens for the base version and $4 and $12 per million Tokens for the high-speed variant. Currently, Cursor and Grok Build users can directly experience the model, and developers can also access it through the API. To celebrate the launch of the new model, Sp