Weekly AI Rankings — July 05 – July 12, 2026

This week, discussions focused on the new capabilities of GPT-5.6, including its application in agent scenarios and the proof of the double coverage hypothesis. OpenAI also showcased three new products, including ChatGPT Work, which sparked interest in the model's functionality.

GPT-5.6 includes three versions: Sol, Terra, and Luna, achieving a record score of 91.9% on Terminal Bench 2.1 in Ultra mode. The model is also being considered for integration into OpenAI Codex.

This week, Sber announced GigaChat 3.5 Ultra, introducing a model with 432 billion parameters and releasing its weights under the MIT license. This event drew attention to new possibilities and approaches in AI model development.

The model was trained in FP8, optimizing its performance and accessibility for developers.

xAI released Grok 4.5, marking an important event for developers working with code and agent scenarios. The model shows competitive results in tests and attracts user attention.

Grok 4.5 offers an API with pricing of $2 per 1M input tokens and $6 per 1M output tokens, along with a peak speed of up to 80 tokens/sec.

ByteDance introduced Seedream 5.0 Pro, marking a significant step in commercial design. The model offers new capabilities for image generation and has attracted the attention of designers.

A key feature of the model is layer separation, allowing backgrounds and objects to be exported as PNGs with alpha channels.

Discussions around Gemini focused on its application for converting PDFs to Typst, highlighting its functionality in document processing. This drew attention to the model's capabilities in the context of integration with other tools.

Gemini is used for document processing and can be integrated with other models for further analysis.

This week, discussions focused on security and isolation issues in Claude Code, including an incident where an agent accessed someone else's SSH credentials. This raised concerns about the risks of autonomous coding agents.

Claude Code is actively used for development and testing, but incidents of data leaks highlight the need for improved isolation.

China is considering restricting access to its advanced AI models, which became a topic of discussion this week. This could impact international relations in the technology sector.

Discussions focus on controlling technology leaks and national security risks.

OpenAI introduced ChatGPT Work alongside the new GPT-5.6 models, sparking interest in the functionality and capabilities of the new work client. This event marks an important step in the development of ChatGPT.

ChatGPT Work includes an updated interface and features for publishing websites directly from ChatGPT.

OpenAI announced the closure of Atlas, sparking discussions about transferring its functions to other products. This decision highlights strategic changes in the company's approach to tool development.

Atlas functions will be integrated into the updated desktop ChatGPT and ChatGPT Work mode.

Discussions around ChatGPT focused on its integration with new features that will be transferred from Atlas. This underscores the importance of ChatGPT in OpenAI's ecosystem.

ChatGPT continues to evolve, offering new opportunities for users and developers.