SpaceXAI Releases Grok 4.5, a Cursor-Trained Model for Coding, Agentic Tasks, and Knowledge Work at $2/M Input

Editor's take

SpaceXAI has introduced Grok 4.5, a new model specifically optimized for coding, complex agentic operations, and knowledge-intensive tasks. This release signifies a strategic move to leverage Cursor's training methodology, aiming for enhanced performance and efficiency in specialized AI applications.

The significance lies in Grok 4.5's purported benchmark performance, topping Harvey's Legal Agent Benchmark, suggesting a strong capability in structured, domain-specific reasoning. Its competitive pricing of $2/$6 per million input/output tokens, coupled with an 80 transactions per second throughput, positions it as a potentially cost-effective solution for businesses seeking advanced AI agents and coding assistants, directly competing with established players like OpenAI's GPT-4.

Future observation should focus on independent verification of its benchmark claims, particularly in real-world coding and agentic task execution beyond legal benchmarks. The long-term impact will depend on its ability to maintain this performance edge at scale and whether this Cursor-trained architecture can be widely adopted for other complex AI workloads, potentially influencing the development trajectory of other large language models.