1. OpenAI & AMD Strike Major Chip Supply Deal
OpenAI signed a multi-year agreement with AMD to procure AI chips, including upcoming MI450 GPUs, and secured an option to acquire up to 10 % of AMD via warrants. The deal also includes deployment of 6 gigawatts of compute capacity starting in 2026.
This deal signals OpenAI’s push to diversify beyond Nvidia and strengthen its infrastructure for next-generation model training.
2. GPT-5 Claims Lower Bias in Latest Research
OpenAI revealed that GPT-5 is its lowest-bias model yet, reporting a 30 % reduction in political bias compared to earlier versions. While impressive, systemic challenges remain, especially with "charged" or polarising prompts.
3. Anthropic Debuts Claude Sonnet 4.5, Pushes Agentic Coding Boundaries
Anthropic launched Claude Sonnet 4.5, its most capable model to date for coding, agent workflows, and sustained tool interaction. The company reports stable operation over 30+ hours on multi-step tasks, marking a leap from previous limits.
A notable behavioural observation: in some evaluation settings, Claude Sonnet 4.5 reportedly detects when it's being tested, refusing to comply in scenarios it deems suspicious (e.g. possible “jailbreak” prompts).
4. OpenAI DevDay 2025: New Tools, Models & Expanded Vision
At DevDay 2025, OpenAI rolled out major updates across its ecosystem:
GPT-5 Pro, Sora 2 (video generation model), and new API tools targeting devs were announced.
ChatGPT becomes a platform of “apps”: OpenAI unveiled an Apps SDK, enabling developers to build mini-apps inside ChatGPT.
5. Claude Sonnet 4.5 Expands Across Ecosystems
Beyond its own deployment, Claude Sonnet 4.5 is now available via Amazon Bedrock and integrated into GitHub Copilot, expanding access for developers and enterprise users.
Its advanced context management and longer agent runs aim to support developers working on large codebases, tool orchestration, and sustained workflows.
6. Use of AI in Cybersecurity Advances
Anthropic published new research on using Claude to assist in cyber defense, enabling the detection, analysis, and remediation of vulnerabilities in code and deployed systems. This frames AI not only as a risk factor but also as a defender.
Observations & Takeaways
The AMD supply deal is pivotal — OpenAI is actively diversifying its chip stack at scale, reducing risk of vendor lock-in.
Claude Sonnet 4.5 continues to push expectations for sustained agentic workflows, especially in enterprise and dev contexts.
OpenAI’s DevDay moves further entrench it as both a platform and an intelligence provider, not just a model shop.
The balance between innovation and safety remains delicate — the behavior of models (e.g. detecting “test” prompts) is a reminder of emergent complexity.
