Last month, we launched Gemma 3, our latest generation of open models. Delivering state-of-the-art performance, Gemma 3 quickly established itself as a leading model capable of running on a single high-end GPU like the NVIDIA H100 using its native BFloat16 (BF16) precision.

To make Gemma 3 even more accessible, we are announcing new versions optimized with Quantization-Aware Training (QAT) that dramatically reduces memory requirements while maintaining high quality. This enables you to run powerful models like Gemma 3 27B locally on consumer-grade GPUs like the NVIDIA RTX 3090.

Today we are rolling out an early version of Gemini 2.5 Flash in preview through the Gemini API via Google AI Studio and Vertex AI. Building upon the popular foundation of 2.0 Flash, this new version delivers a major upgrade in reasoning capabilities, while still prioritizing speed and cost. Gemini 2.5 Flash is our first fully hybrid reasoning model, giving developers the ability to turn thinking on or off. The model also allows developers to set thinking budgets to find the right tradeoff between quality, cost, and latency. Even with thinking off, developers can maintain the fast speeds of 2.0 Flash, and improve performance.

IBM (NYSE: IBM) today released the 2025 X-Force Threat Intelligence Index highlighting that cybercriminals continued to pivot to stealthier tactics, with lower-profile credential theft spiking, while ransomware attacks on enterprises declined. IBM X-Force observed an 84% increase in emails delivering infostealers in 2024 compared to the prior year, a method threat actors relied heavily on to scale identity attacks.

The 2025 report tracks new and existing trends and attack patterns – pulling from incident response engagements, dark web and other threat intelligence sources.

AI is moving faster—and getting smarter. Today’s open large language  models are not only powerful but also designed with real-world  deployment in mind: they’re lightweight, cost-efficient, and built to  scale across billions of devices. In short, they’re ready for just about  anything developers can imagine.

In the fifth episode of Tech Unheard, Scale AI CEO Alex Wang joins  Arm CEO Rene Haas to discuss the connections between humans, data and AI  – as part of what Alex calls “humanity-first AI.”

Alex founded Scale AI in 2016 at the age of 19, with the company now working with Microsoft, Meta, OpenAI and more. He talks with Rene about why youth could be an advantage in the world of technology, while reflecting on his own upbringing among scientists near Los Alamos  National Laboratory and learning how to manage stress through math competitions.

Today, we’re releasing OpenAI o3 and o4-mini, the latest in our o-series of models trained to think for longer before  responding. These are the smartest models we’ve released to date,  representing a step change in ChatGPT's capabilities for everyone from  curious users to advanced researchers. For the first time, our reasoning  models can agentically use and combine every tool within ChatGPT—this  includes searching the web, analyzing uploaded files and other data with  Python, reasoning deeply about visual inputs, and even generating  images. Critically, these models are trained to reason about when and  how to use tools to produce detailed and thoughtful answers in the right  output formats, typically in under a minute, to solve more complex  problems. This allows them to tackle multi-faceted questions more  effectively, a step toward a more agentic ChatGPT that can independently  execute tasks on your behalf. The combined power of state-of-the-art  reasoning with full tool access translates into significantly stronger  performance across academic benchmarks and real-world tasks, setting a  new standard in both intelligence and usefulness.

As Mark Hambleton, Arm’s SVP of Software, says in the Arm Silicon Reimagined report: “The future of AI development relies on the synergy between software and hardware.”

However, the big challenge, as set out in a new Arm-sponsored CIO report,  is that developer workflows are often fragmented. This means developers are unable to move as quickly as they would like when creating and scaling new AI applications.

As AI drives demand for advanced logic and memory technologies, the chipmaking roadmap is becoming more complex – opening the door to a growing number of defects that can dramatically impact chip performance and yield. During the 2025 SPIE Advanced Lithography + Patterning  conference, Applied Materials convened experts from IBM, imec, Intel, and Samsung to discuss how advancements in process control can help address emerging defectivity challenges in the “angstrom era.”

Broadcom Inc. (NASDAQ:AVGO) today announced Incident Prediction, an industry-first security capability that extends Adaptive Protection, a unique feature of Symantec Endpoint Security Complete (SES-C), by leveraging AI to identify and disrupt living-off-the land (LOTL) attacks and other cyberthreats.

Texas Instruments (TI) (Nasdaq: TXN) today introduced a new portfolio of automotive lidar, clock and radar chips to help automakers transform vehicle safety by bringing more autonomous features to a wider range of cars. TI's new LMH13000, the industry's first integrated high-speed lidar laser driver, delivers ultra-fast rise time to improve real-time decision-making. The industry's first automotive BAW-based clocks, the CDC6C-Q1 oscillator and LMK3H0102-Q1 and LMK3C0105-Q1 clock generators, improve advanced driver assistance system (ADAS) reliability. Addressing evolving ADAS needs, TI's new AWR2944P mmWave radar sensor offers advanced front and corner radar capabilities.

Today, we’re launching three new models  in the API: GPT‑4.1, GPT‑4.1 mini, and GPT‑4.1 nano. These models  outperform GPT‑4o and GPT‑4o mini across the board, with major gains in  coding and instruction following. They also have larger context  windows—supporting up to 1 million tokens of context—and are able to  better use that context with improved long-context comprehension. They feature a refreshed knowledge cutoff of June 2024.