SpaceXAI introduced Grok 4.5: an updated model for programming and agent tasks

On July 8, 2026, SpaceXAI announced the release of a new version of its Grok 4.5 language model. The update is focused on practical applications in software development: from code generation and refactoring to performing multi-step tasks in AI agent mode. The model is already available via API and integrated into developer tools, including Cursor, allowing it to be used in familiar workflows. According to the company, Grok 4.5 was trained on an expanded set of engineering problems and received improvements in contextual understanding, working with multi-file projects and handling long queries, xrust reports. SpaceXAI notes that the model is able to analyze the structure of the project, suggest correction options and bring the solution to a working state without the need to manually refine each step. This makes it popular in scenarios where not only code generation is required, but also sequential execution of actions.

Russian developers are already testing the model in the Cursor editor and through the API. In specialized communities, they note that Grok 4.5 better copes with tasks related to legacy code support, integrations and automation of routine processes. In these areas, not only algorithms are important, but also the ability of the model to work correctly with heterogeneous projects where careful changes are required.

The cost of use attracts special attention. According to SpaceXAI, the price is about 153 rubles per million input tokens and approximately 460 rubles per million output tokens when converted at the current rate. Compared to competitors, this looks moderate, especially for small teams and freelancers. In discussions, the developers note that the model economically uses tokens in engineering tasks, which reduces the final costs. At the same time, the company does not position Grok 4.5 as an absolute leader in quality: in independent tests it is inferior to the flagship OpenAI and Anthropic models, but shows stable results in application scenarios.

In the Russian context, interest in Grok 4.5 is associated with several factors. First, developers are looking for tools that can speed up work without significantly increasing costs. Secondly, there is a growing demand for models that can work in agent mode: perform chains of actions, generate documentation, prepare presentations and automate testing. Thirdly, many teams continue to use foreign AI services through APIs, despite restrictions on regional availability. Grok 4.5 has no official restrictions for Russia, which allows you to integrate the model into workflows without additional technical workarounds.

Experts note that the emergence of new models with a focus on engineering tasks is increasing competition in the AI ​​tools market. For developers, this means expanded choice and the ability to select solutions for specific tasks. Discussions on Habré and in Telegram channels provide the first examples of using Grok 4.5: generating template code, automating tests, preparing technical descriptions and working with large documents. Users emphasize that the model copes better with sequential tasks than previous versions.

SpaceXAI continues to actively develop the Grok line. The company announces plans to regularly update and expand the functionality of its models, including improving performance in agent scenarios. For the market, this means further movement towards tools that not only generate text, but also perform actions, moving closer to full-fledged digital assistants for developers.

In general, the release of Grok 4.5 reflects the current trend in the industry: attention is shifting from universal models to specialized solutions focused on specific professional tasks. For Russian developers, this is another tool that can increase work efficiency and reduce time for performing routine operations. At the same time, experts emphasize that the use of AI does not eliminate the need for quality control: the final code must be checked by an engineer, especially in projects with increased reliability requirements.

Sources: Official announcement of SpaceXAI, Reuters, Discussion on Habré

Xrust SpaceXAI presented Grok 4.5: updated model for programming and agent tasks