August was mainly marked by releases from OpenAI, but much more happened. We saw the introduction of GPT-5, open-source models and the new Codex IDE. Anthropic introduced the millionth context and Google showed two new models for working with images.
GPT-5
OpenAI has launched a new generation of its GPT-5 model. What's new?
- Multiple versions for different needs: standard, mini and nano. Mini and nano versions are optimized for speed and price, standard offers balanced performance.
- Cheaper operation: router distributes queries between models → more efficient use of resources.
- Less accommodating responses: the GPT-5 minimizes over-accommodating "sycophancy", improves the interpretation of advanced questions and better adapts the communication style.
- Focus on technical roles: better results in technical and programming tasks, even though solutions may take longer.
- Integration: access to external services (GitHub, Google Drive, Gmail, Calendar) for Plus users.
- Reasoning effort: the ability to set how "deeply" the model should reason.
- Verbosity: checking the length and detail of the answer (concise vs. explanatory).
- Custom tools: tools defined in plain text or using context-free grammar → more flexible work with tool calls.
GPT OSS
OpenAI also introduced open source models
- two models with 20M and 120M parameters, they have quite different architecture
- given their size, both achieve very good results
- use s → only part of the model is activated when queried
- emphasis on sustainability and resource optimization
GPT Real time
The next release from OpenAI is a model optimized for live scenarios, especially call centers.
- speech-to-speech
- also handles video inputs
- can call tools using MCP
OpenAI Codex IDE
OpenAI is expanding its portfolio for developers as well.
- new coding assistant directly in the IDE
- available under the OpenAI Single License
- handles full agent mode for vibe coding, like Claude Code
Claude 4 Sonnet
Anthropic has come up with a major shift: the Claude 4 Sonnet now handles 1 million tokens in context. This means the ability to work with a huge amount of text at once. This is also reflected in the price.
Gemini 2.5. Flash Image
Google is coming up with a new model for image editing. The model is already a few days ahead of its official release on one of the community platforms under the name Nano Banana.
- it mainly focuses on photo editing, for example, to color a black and white image or calculate how the object would look from behind
- convert 2D to 3D, merge multiple images or add realistic shadows
- he is not very good at text in pictures, but otherwise he is one of the best we have seen so far
We attach an example of his work, with a prompt: "This guy sitting with that dog on the sofa in the same room."
Claude for Chrome
Anthropic is testing a Chrome extension:
- agent assistant directly in the browser
- can order goods, manage emails and interact with the web
August brought major news mainly from OpenAI - GPT-5, open-source models and new IDEs. Anthropic moved to million context and Google moved to more realistic image handling. The common denominator is more emphasis on practical use and more control for developers, who now get tools better suited to real-world scenarios.
Author
Jakub Vacek
Software AI EngineerSenior backend developer specializing in microservices architecture, with a growing focus on AI. Skilled in TypeScript, Node.js, React.js, Nest.js, and building AI-powered solutions.
