AI Tech Weekly

**OpenAI’s Shift Toward Open-Weight Models and Next-Generation LLMs**
OpenAI is preparing to release its first open-weight language model since GPT-2 (2019), codenamed “o3-mini,” as early as the coming week. Unlike its proprietary GPT-4 and GPT-4o models, this open-weight release will allow third parties—governments, research labs, and companies—to run the model independently of cloud providers. It is expected to be available on platforms like Microsoft Azure and Hugging Face, challenging the exclusivity of OpenAI’s partnership with Microsoft, which currently enjoys a 20% revenue share from both ChatGPT subscriptions and the Azure OpenAI Service. The move raises questions about licensing terms and the degree of openness, given Microsoft’s limited control over an open model’s deployment and use. CEO Sam Altman has hinted at the arrival of an open model “this month,” signaling OpenAI’s strategic pivot toward greater accessibility in AI research and deployment ([theverge.com](https://www.theverge.com/notepad-microsoft-newsletter/702848/openai-open-language-model-o3-mini-notepad?utm_source=chatgpt.com), [gadgets360.com](https://www.gadgets360.com/ai/news/openai-gpt-5-o-series-unified-ai-model-new-capabilites-features-release-8835969?utm_source=chatgpt.com)).

In parallel, OpenAI executives have teased GPT-5 as a unification of the GPT-series and the o-series (omni) models. Romain Huet, Head of Developer Experience, described GPT-5 as combining the advanced reasoning capabilities of o-series with the multimodal strengths of GPT-series, effectively creating a single, “great frontier model.” Although no concrete release date has been announced, Sam Altman has suggested in public streams that GPT-5 could debut this summer, potentially in July or August. Anticipated features include dramatic context-length expansion, improved reasoning to reduce hallucinations, deeper code comprehension, and enhanced multimodal inputs for images and video. This roadmap follows OpenAI’s February unveiling of GPT-4.5 as a research preview, marking incremental improvements in natural dialogue and “EQ” before the major leap promised by GPT-5 ([gadgets360.com](https://www.gadgets360.com/ai/news/openai-gpt-5-o-series-unified-ai-model-new-capabilites-features-release-8835969?utm_source=chatgpt.com), [tomsguide.com](https://www.tomsguide.com/ai/gpt-5-could-be-here-this-month-heres-five-features-were-hoping-to-see?utm_source=chatgpt.com)).

**Lifecycle Transitions: Deprecation of Older Models**
As OpenAI prepares new model releases, it is concurrently phasing out legacy offerings. According to GitHub’s changelog and OpenAI announcements, GPT-4.5 will be removed from GitHub Copilot and Copilot IDE model pickers on July 7, 2025, with GPT-4.1 designated as the default replacement. The GPT-4.5 Preview API is set for deprecation on July 14, 2025. Similarly, GPT-4o—the multimodal version of GPT-4—and initial o-series models like o1 and o3-mini are slated for removal around the same period, with successors such as o3 and o4-mini recommended. OpenAI’s rationale centers on optimizing performance, reducing computational costs, and encouraging developers to adopt more capable models offering features like up to one million token contexts compared to GPT-4o’s 128,000-token limit ([poniaktimes.com](https://www.poniaktimes.com/openai-model-deprecations-2025/?utm_source=chatgpt.com), [help.openai.com](https://help.openai.com/en/articles/9624314-model-release-notes/?utm_source=chatgpt.com)).

**Regulatory Milestones: EU Code of Practice and U.S. State-Level Legislation**
On July 10, 2025, the European Commission published the final version of its voluntary Code of Practice for general-purpose AI, designed to guide companies in complying with the EU AI Act ahead of enforcement starting August 2. Developed by 13 independent experts with contributions from over 1,000 stakeholders—including SMEs, academics, and civil society—the code covers transparency, copyright, safety, and systemic risk management. While compliance remains voluntary, signatories can benefit from reduced administrative burdens and clearer legal frameworks. Non-compliance under the AI Act can carry fines up to €35 million or 7% of global revenue. Major AI providers such as OpenAI and Google are actively reviewing the code, even as some industry leaders call for further simplification to balance innovation and regulatory clarity ([wsj.com](https://www.wsj.com/tech/ai/eu-lays-out-voluntary-ai-code-of-practice-to-guide-companies-on-compliance-638497a8?utm_source=chatgpt.com), [pymnts.com](https://www.pymnts.com/news/artificial-intelligence/2025/european-union-publishes-final-ai-code-guide-compliance-tech-companies/?utm_source=chatgpt.com)).

Meanwhile, at the state level in the U.S., Texas lawmakers passed one of the most comprehensive AI bills in the country, set to take effect in 2026. The legislation mandates transparency requirements for automated decision-making, bias mitigation protocols, and a framework for periodic AI audits. It applies to both government and private-sector uses of AI technologies, positioning Texas as a leader in proactive AI governance. Officials contend the law strikes a balance between fostering technological innovation and protecting citizens from algorithmic harms, reflecting a growing trend of state-level AI regulation in the United States ([crescendo.ai](https://www.crescendo.ai/news/latest-ai-news-and-updates?utm_source=chatgpt.com)).

**Hardware and Infrastructure: New GPU Solutions and Energy Strategies**
Sparkle has launched its Arc Pro B60 series of professional GPUs based on Intel’s Battlemage architecture, featuring single-GPU and dual-GPU cards with up to 48 GB of GDDR6 memory per board. Targeted at AI inference workloads and professional visualization, the high-end dual-GPU variant delivers up to 394 TOPS of INT8 performance and includes full Linux multi-GPU support via Intel’s OneAPI. By focusing on higher memory capacities and optimizations for AI tasks, the Arc Pro B60 positions itself as a cost-effective alternative to leading Nvidia and AMD professional GPUs, especially for enterprises seeking diverse compute options beyond traditional vendors ([techradar.com](https://www.techradar.com/pro/another-dual-intel-gpu-card-with-48gb-vram-launched-as-arc-pro-emerges-as-cheap-nvidia-and-amd-alternative-for-ai-crowd?utm_source=chatgpt.com)).

On the energy front, Elon Musk’s xAI announced plans to build a dedicated overseas power plant to run a planned fleet of one million GPUs. The initiative aims to circumvent the limitations of U.S. grid infrastructure by constructing a self-contained energy facility to support large-scale model training without impacting local power stability. Musk cited current constraints on power availability and the need for sustainable, high-capacity compute environments as driving factors. This project underscores the escalating energy demands of frontier AI research and the strategic infrastructural investments required to sustain it ([crescendo.ai](https://www.crescendo.ai/news/latest-ai-news-and-updates?utm_source=chatgpt.com)).

**Healthcare Innovations: AI-Designed Drugs and Immunotherapy Collaborations**
On July 7, 2025, DeepMind spinoff Isomorphic Labs announced that its AI-designed therapeutic candidates are entering human trials, marking a significant milestone in AI-driven drug discovery. By leveraging machine learning to model protein interactions and predict compound efficacy, Isomorphic Labs claims to accelerate the identification of viable drug candidates, potentially reducing traditional discovery timelines from years to months. The technology promises to enhance precision in targeting diseases such as cancer and neurological disorders by surfacing novel molecules with optimized interaction profiles ([crescendo.ai](https://www.crescendo.ai/news/ai-in-healthcare-news?utm_source=chatgpt.com)).

In parallel, Summit Therapeutics and Akeso Inc. unveiled a $5 billion+ licensing and collaboration deal to co-develop the PD-1/VEGF bispecific antibody Ivonescimab using AI-driven immunotherapy platforms. Summit will apply AI analytics to biomarker identification and trial design, aiming to streamline patient stratification and endpoint optimization. The partnership targets non-small cell lung cancer and other solid tumors, showcasing AI’s expanding role in complex biologics development and personalized oncology treatments ([crescendo.ai](https://www.crescendo.ai/news/ai-in-healthcare-news?utm_source=chatgpt.com)).

**Institutional and Industry Applications: Standards and Global Governance**
In healthcare standards development, HL7 International officially launched its dedicated AI Office on July 8, headquartered in Ann Arbor, Michigan, under the leadership of Daniel Vreeman as Chief AI Officer. The office will spearhead the HL7 International AI Challenge (open through July 30) and work closely with regulators and industry partners to establish interoperable, explainable, and scalable AI frameworks for clinical, operational, and research applications. This move reinforces the central role of standards bodies in ensuring trust and consistency as AI permeates healthcare ecosystems ([aiin.healthcare](https://aiin.healthcare/newsletters?utm_source=chatgpt.com)).

Beyond individual sectors, BRICS nations proposed on July 7 that the United Nations take the lead in forming global AI governance frameworks. Arguing that current norms are overly influenced by Western technology firms, the bloc seeks equitable representation and oversight to ensure developing economies benefit from AI advances. The proposal, submitted to the UN secretary-general, could influence the formation of international standards addressing ethical, safety, and access considerations, reflecting a shift toward multipolar AI governance dialogues ([crescendo.ai](https://www.crescendo.ai/news/latest-ai-news-and-updates?utm_source=chatgpt.com)).

Moreover, major insurance companies are rapidly embedding AI into underwriting, claims processing, and fraud detection. While executives celebrate improved efficiency and risk assessment accuracy, they express concern over regulatory uncertainty and the potential for algorithmic bias. Industry leaders are advocating for clear federal guidelines to navigate evolving compliance landscapes, highlighting the interplay between AI-driven operational gains and the need for robust governance to maintain consumer trust ([crescendo.ai](https://www.crescendo.ai/news/latest-ai-news-and-updates?utm_source=chatgpt.com)).

—
This overview captures the breadth of AI developments from July 7 to 13, 2025, spanning model releases and deprecations, regulatory milestones in Europe and the U.S., advancements in hardware and infrastructure, breakthroughs in healthcare applications, and emerging standards and governance initiatives.