XTHE

Microsoft Copilot for GCC High, released in late 2025, brings AI features created for defense contractors and government agencies. It focuses on data sovereignty. Unlike the more feature-rich and web-connected commercial Copilot, GCC High runs in a separate environment with US-based data storage and vetted staff web grounding. It is usually turned off by default to help prevent data leaks.  

Main differences: Copilot GCC High vs Commercial 

• Security and compliance: GCC High meets stringent standards, including DFARS, ITAR, and FedRAMP High. All data is saved and processed in the US Sovereign Cloud. In contrast, commercial users use public cloud infrastructure.  

• Web grounding (search): In GCC High, web grounding is off by default to keep sensitive data within the compliance boundary. Commercial allows full web access for the latest information.  

• Feature Availability and Timing: Commercial gets new features first. GCC High may receive some features later, such as Copilotgraph grounding, due to strict review processes.  

• Target audience: GCC High is made for the defense of industrial base and government agencies. Commercial is for all other businesses.  

• Data access: Both versions use Microsoft Graph, but GCC High keeps data within the compliant boundary.  

Summary Table 

Feature	Microsoft Copilot(Commercial)	Microsoft Copilot(GCC High)
Cloud Environment	Public Azure	Isolated Azure Government
Data Sovereignty	US-based (primarily)	Strict U.S. only
Compliance	Iso/soc/HIPAA/GDPR	DFARS, ITAR FedRamp high
Web Grounding	Enabled by default	Disabled by default (opt in)
Future Release	First	Delayed/Phased

Important Points for GCC High 

• Oversharing risks: Organizations need to set up strong data access controls. Copilot can disclose sensitive files if they are overshared, which is especially risky in high-compliance settings.  

• Policy management: GCC lets organizations decide if and how users can access online web data.  

• Deployment: GCC High generally became available in December 2025.  

AI You Can Trust, Designed To Give You Confidence 

Microsoft 365 Copilot brings together advanced language models and Work IQ, an intelligence layer that helps CoPilot understand you, your work, and your organization. This understanding is built into familiar Microsoft 365 apps like Word, Excel, PowerPoint, and Outlook, helping you work more efficiently.  

With Microsoft 365 Copilot, agencies can make citizen services more efficient by quickly drafting responses and summarizing case files. Staff can also manage budgets more effectively by analyzing spending trends and creating reports that meet compliance standards, all within the Microsoft 365 apps they already use.  

Copilot for GCC High is designed to meet the strict regulations that many agencies and related organizations must follow. It complies with standards such as FedRAMP High, DFARS, ITAR, CMMC, and other key requirements. Key protections include:  

• Data Residency and Isolation: All data stays in US-based data centers managed by approved US staff meeting government requirements.  

• Encryption and access controls: data is secured both during transmission and storage. Microsoft Entra ID manages who can access the data based on their role.  

• Responsible AI by design: Copilot follows Microsoft’s responsible AI principles and includes protections against prompt injection and misuse.  

To keep sensitive government data within the GCC High compliance boundary, Microsoft 365 Copilot comes with Web Grounding turned off by default. Turning this setting on when needed can improve Copilot’s responses. For more information, visit Data Privacy and Security for Microsoft 365 Copilot.

Source: Microsoft 365 Copilot is now available in GCC-High

NVIDIA introduced the Blackwell Ultra B300 Data Center GPU today at CEO Jensen Huang’s GTC 2025 keynote in San Jose, CA. The new GPU has 50% more memory and FP4 compute than the B200, pushing the competition for faster, more advanced AI models even further. NVIDIA describes it as built for the age of reasoning, pointing to advanced AI LLMs like DeepSeek R1 that can do more than regurgitate information they’ve already processed.  

The Blackwell Ultra B300 is far more than a single GPU. In addition to the base B300, NVIDIA is releasing the new B300 NVL16 server racks, a GB300 DGX station, and GB300 NV72L full rack solutions. Combining 8 NV72L racks creates the Blackwell Ultra DGX SuperPod, which includes:  

• 288 GRES CPUs  

• 576 Blackwell Ultra GPUs  

• 300 TB of HBM3e memory  

• 11.5 EXA flops of FP4  

NVIDIA refers to these large systems as AI factories.  

NVIDIA claims the Blackwell Ultra will have 1.5 times the FP4 compute density, but it is unclear whether other compute types have increased by the same amount. We expect similar improvements, but NVIDIA may have made changes beyond just adding more SMs, such as raising clock speeds or expanding HBM3e memory. For example, clock speeds are lower in FP8 or FP16 modes. Here are the main specs we know so far, with some inferred data indicated by question marks.  

NVIDIA Blackwell Ultra B300 vs Blackwell B200 

Platform	B300	B200	B100
Configuration	Blackwell GPU	Blackwell GPU	Blackwell GPU
FP4 Tensor Dense/Sparse	15/30 Petaflops	10/20 Petaflops	7/14 Petaflops
FB6/FB8 tensorDense/Sparse	7.5/15 petaflops?	5/10 petaflops	3.5/7 petaflops
INT 8 Tensor Dense/Sparse	7.5/15 petaops?	5/10 petaops	3.5/7 petaops
FP16/BF16 Tensor Dense/ sparse	3.75/7.5 Petaflops?	2.5/5 Petaflops	1.8/3.5 petaflops
TF32 Tensor Dense/Sparse	1.88/3.75 petaflops?	1.25/2.5 Petaflops	0.9/1.8 petaflops
FP-64 Tensor Dense	68 TeraFLOPS?	45 TeraFLOPs.	30 teraflops
Memory	288 GB (8x 36 GB)	192 GB (8x 24 GB)	192 GB (8x 24 GB)
Bandwidth	8 TB/s?	8 TB/s	8 TB/s
Power	?	1300 W	700 W

Asked NVIDIA for more details about the Blackwell Ultra B300’s performance and received this response: “Blackwell Ultra GPUs in GB300 and B300 are different chips than Blackwell GPUs in GB200. Blackwell Ultra GPUs are designed to meet the demand for test-time scaling inference with a 1.5x increase in FP4 compute. This suggests that the B300 might be a physically larger chip to accommodate more tensor scores, but we are still waiting to verify.”  

The new B300 GPUs will deliver much higher computing throughput than the B200, with 50% more on-package memory. They can support even larger AI models with more parameters, and the extra compute power will be a big advantage.  

NVIDIA shared some performance examples, but compared the B300 to Hopper, which makes the results less clear. It would be more helpful to see direct comparisons between the B200 and B300 using the same number of GPUs, but that information isn’t available yet.  

With FP4 instructions and the new Dynamo software library, the B300 can serve reasoning models like DeepSeek much more efficiently. NVIDIA claims an NV72L rack can deliver 30 times the inference performance of a similar Hopper setup. This improvement comes from faster NVLink, more memory, increased compute, and the use of FP4.  

For example, Blackwell Ultra can process up to 1,000 tokens per second with the DeepSeek R1671B model, whereas Hopper can process only 100 tokens per second. This means throughput is ten times higher, reducing the time to handle a large query from 1.5 minutes to just 10 seconds.  

300 products are expected to start shipping in the second half of the year before year-end, and NVIDIA hopes to avoid any packaging issues or delays this time. Last fiscal year, the company made $11 billion from Blackwell B200/B100, and it’s probably targeting an even higher number this year.

Source: Nvidia announces Blackwell Ultra B300 —1.5X faster than B200 with 288GB HBM3e and 15 PFLOPS dense FP4 

We are excited to introduce GPT-5.3 Codex, our most advanced agentic coding model yet. It combines the top coding performance of GPT-5.2 Codex with the reasoning and professional knowledge of GPT-5.2, all in a model that runs 25% faster. This means it can handle long tasks like research and tool use, as well as complex projects. You can guide and interact with GPT-5.3 Codex as it works, just like you would with a colleague, and it keeps track of the context throughout.  

GPT-5.3 Codex is our first model instrumental in its own creation. The Codex team used early versions to debug its own training, manage its own deployment, and analyze test results and evaluations. Our team was blown away by how much Codex accelerated its own development.  

With GPT-5.3 codecs, move beyond just writing and reviewing code. Now it can handle almost anything developers and professionals do on a computer.  

Frontier Agentic Capabilities 

GPT-5.3 Codex sets new records on SWE Bench Pro and Terminal Bench, and also performs well on OS World and GDPval. These four benchmarks help us measure coding agent tech and everyday skills.  

Coding 

GPT-5.3 Codex delivers top results on SWE Bench Pro, a tough test for actual software engineering. Unlike SWE Bench, which covers only Python, SWE Bench Pro includes four languages and is more challenging and industry relevant. GPT-5.3 Codex also beats the previous best on Terminal Bench 2.0, which tests terminal skills. It achieves all this using fewer tokens than any earlier model, so that users can do even more.  

Web Development 

With its improved coding skills, better design, and efficient resource use, GPT-5.3 Codex can build complex games and apps from scratch in just a few days to test its web development and long-running abilities. We asked it to create two games:  

• An updated racing game from the Codex app launch  

• A new diving game using its web game development skills and straightforward follow-up prompts like “fix the bug” or “improve the game.”  

GPT-5.3 Codex worked on the games by itself over millions of tokens. You can watch the trailers and try the games to see what Codex can do.  

GPT-5.3 Codex is also better at understanding what you want when you ask it to create everyday websites compared to GPT-5.2 Codex. Even with simple or vague prompts, it now builds sites with more features and smart defaults, giving you a better starting point for your ideas.  

For example, we asked GPT-5.3 Codex and GPT-5.2 Codex to build two landing pages. GPT-5.3 Codex automatically displayed the yearly plan as a discounted monthly price, making the discount feel clear and intentional. Rather than multiplying the yearly total, it also created an automatically transitioning testimonial carousel with three distinct user quotes, resulting in a page that feels more complete and production-ready by default.  

Beyond Coding 

Software engineers, designers, product managers, and data scientists do much more than write code. GPT-5.3 Codex is designed to support every part of the software lifecycle, including debugging, deploying, monitoring, and writing PRDs. Its capabilities also go beyond software, so you can use it to create slide decks or analyze data within spreadsheets.  

Using custom skills like those from our earlier GDPval results, GPT-5.3 Codex also performs well on professional knowledge work, matching GPT-5.2. GDP Well is an evaluation released by OpenAI in 2025 that measures how well a model handles specific knowledge work tasks across 44 jobs. These tasks include making presentations, creating spreadsheets, and producing other work products.  

These results across coding, front-end, computer use, and real-life tasks show that GPT-5.3 Codex is not just better at individual tasks. It represents a major move toward a single, general-purpose agent that can reason, build, and execute across a wide range of technical work.  

An Interactive Collaborator 

As models become more powerful, the main challenge is making it easy for people to interact with many agents simultaneously. The Codex app helps you manage and guide agents more easily, and with GPT-5.3 Codex, it is now more interactive. The new model gives frequent updates so you can keep informed about key decisions and progress. Instead of waiting for a final result, you can interact in real time by asking questions, discussing approaches, and guiding the solution. Codex explains what it is doing, responds to feedback, and keeps you updated from start to finish.  

How We Used Codex To Train And Deploy Gpt-5.3 Codex 

Recent improvements to Codex are the result of research projects at OpenAI and have taken months or even years to develop. Codex is speeding up these projects, and many researchers and engineers at OpenAI say their work now feels very different compared to just two months ago. Even the early versions of GPT-5.3 Codex showed strong abilities, which helped our team improve training and support the launch of later versions. Codex is useful for a wide range of tasks, making it difficult to enumerate all the ways it helps our teams. As some examples, the research team used Codex to monitor and debug the training run for this release. It accelerated research beyond debugging infrastructure problems:  

• It helped monitor column patterns throughout training.  

• Provided a deep analysis of interaction quality.  

• Proposed fixes and built rich applications for human researchers to precisely understand how the model’s behavior differed from prior models.  

The engineering team used Codex to optimize and adapt the harness for GPT-5.3 Codex. When we started seeing strange edge cases affecting users, team members used Codex to identify context-rendering bugs and the root cause of low cache hit rates. GPT-5.3 Codex is continuing to help the team throughout the launch by dynamically scaling graphics processing unit clusters to adjust to traffic surges and keeping latency stable.  

During Alpha testing, a researcher wanted to see how much extra work GPT-5.3 Codex completed per turn and how it affected productivity. GPT-5.3 Codex completed simple regex classifiers to measure how often clarifications were needed to track user responses and monitor task progress. It then ran these checks across session logs and produced a report with its findings. People using Codex were happier because the agent better understood their intent and made more progress each term with fewer explanatory questions.  

Because GPT-5.3 Codex is so different from earlier versions, the alpha testing data showed many unusual and unexpected results. A data scientist on the team worked with GPT-5.3 Codex to build new data pipelines and create better visualizations than our usual dashboard tools. Together, they quickly analyzed the results, and Codex summarized key insights from thousands of data points in less than three minutes. These tasks are interesting examples of how Codex can help researchers and produce builders. Taken together, we found that these new capabilities accelerated our research, engineering, and product teams.  

Securing the Cyber Frontier 

In recent months, we have seen real improvements in model effectiveness on cybersecurity tasks, which helps both developers and security professionals. At the same time, we have been working on stronger cyber safeguards to support defensive use and make the ecosystem more resilient.  

GPT-5.3 Codex is the first model we classify as having high capability for cybersecurity-related tasks under our preparedness framework, and the first we’ve directly trained to identify software vulnerabilities. While we don’t have conclusive evidence that it can automate end-to-end cyber-attacks, we are taking a precautionary approach and deploying our most extensive cybersecurity safety stack to date. Our mitigations include:  

• Safety training  

• Automated monitoring  

• Trusted access for sophisticated capabilities  

• Enforcement pipelines, including threat intelligence  

Since cybersecurity can be used for both good and bad purposes, we are using an evidence-based, step-by-step approach. This helps defenders find and fix vulnerabilities faster while making misuse harder. As part of this, we are launching Trusted Access for Cyber, a pilot program to speed up cyber defense research.  

To help prevent misuse, some requests that our systems see as higher cyber risks may be automatically sent from GPT-5.3 Codex to GPT-5.2. We are still improving these safeguards. Developers doing security research or who think their requests are classified can apply for full access through our trusted access program, or report the issue using the feedback command.  

We are investing in ecosystem safeguards by expanding the private beta of Aardvark, our security research agent. This is the first product in our Codex Security Suite. We are also working with open-source maintainers to offer free codebase scanning for popular projects like Next.js, where a security researcher recently used Codex to find and disclose vulnerabilities.  

Building on our 1 million Cyber Security grant program launched in 2023, we are also committing $10 million in API credits to accelerate cyber defense with our most capable models, especially for open-source software and critical infrastructure systems. Organizations engaged in good faith security research can apply for API credits and support through our Cyber Security grant program.  

Availability & Details 

GPT-5.3 is available with paid ChatGPT plans. Wherever you use Codex: the app, CLI, IDE extension, and web, we are working to enable API access safely in the near future.  

With this update, we are also running GPT-5.3 Codex 25% faster for Codex users, thanks to improvements in our infrastructure and inference stack, resulting in faster exchanges and faster results.  

GPT-5.3 Codex was designed and trained on NVIDIA GB200/NVL72 systems. We thank NVIDIA for its partnership.

Source: Introducing GPT‑5.3‑Codex 

The Samsung Galaxy S26 Ultra and S26 Plus, expected in early 2026, mainly differ in their camera and display features.  

The S26 Ultra features a 6.9-inch screen, a 200MP quad-camera, and S Pen support. It will likely use the faster Snapdragon 8 Elite Gen 5 chip. The S26 Plus comes with a 6.7-inch display and a triple 50 MP camera and usually runs on the Exynos 2600 chip in some areas. Both models offer about 2,600 nits of brightness and support 45 W charging.  

Key Technical Differences 

• Camera System: The S26 Ultra is expected to have a 200MP main camera, a 50MP ultra-wide lens, and a 50MP periscope lens with 5x zoom. The S26+ will likely include a 50MP main camera, a 12MP ultra-wide lens, and a 10MP telephoto lens with 3x zoom.  

• Performance: The S26 Ultra will likely use a faster Snapdragon 8 Elite Gen 5 chip reaching up to 4.74GHz. The S26 Plus is expected to use the Exynos 2600 chip in many regions.  

• The Ultra has a larger 6.9-inch display with possibly higher peak brightness close to 3000 nits and likely supports the S Pen. Both models will feature Gorilla Glass armor and a privacy display.  

• Battery: The S26+ is expected to come with a 4900 mAh battery and 45W charging. The Ultra may have a larger upgraded battery.  

• Design: Both phones are expected to be thinner and lighter, with stronger, more refined frames.  

Summary Table of Expected Specifications 

Feature	Galaxy S26 Ultra	Galaxy S26 Plus
Display	6.9-inch dynamic AMOLED 2x	6.7-inch Dynamic AMOLED 2X
Processor	Snapdragon 8 Elite Gen 5	Exynos 2600/Snapdragon
Rear Camera	200 MP + 50 MP + 50 MP + 10 MP	50 MP + 10 MP + 12 MP
Battery	~5000 Plus MH	4900 mAh
Charging	45 W Plus	45 W Plus
S. Pen	Yes	No.
Protection.	Gorilla Glass Armor	Gorilla Glass Armor

The Galaxy S26 series is set to arrive in less than 2 weeks. Samsung will release the new flagships on Feb 25, 2026, and they’ll be available in stores starting in March.  

This is a little later than expected, but there’s a good reason. Samsung originally planned to launch the Galaxy S26 Pro and Galaxy S26 Edge, but those models were reportedly canceled, so the company returned to its usual lineup.  

As a result, we’ll see the familiar lineup: the Galaxy S26, Galaxy S26 Plus, and Galaxy S26 Ultra.  

As usual, the Galaxy S26 Plus and Galaxy S26 Ultra will be the larger models. How do they stack up against each other?  

Design and Size 

Samsung is sticking to a safe approach. Again, this year, neither the Galaxy S26 Plus nor the Galaxy S26 Ultra will have major design changes. All updates will likely be small and won’t affect how the phones feel to use.  

The Galaxy S26 Plus will likely retain the design style of the previous Galaxy S Plus models. It should be a compact and slim phone with well-placed buttons. The phone is expected to keep the aluminum flat frame and flat Gorilla Glass Victus 2 2 panels on both the front and the back.  

The main difference from the previous generation might be a slightly raised camera island on the back, which will hold all three camera lenses. The Galaxy S25 Plus did not have this kind of unified camera island.  

The Galaxy S26 Ultra will likely use aluminum rather than titanium, as Apple did with the iPhone 17 series. Aside from that and a slightly raised camera island on the back, the Galaxy S26 Ultra will look much like the Galaxy S25 Ultra.  

The Galaxy S26 Plus will be the more compact option. The Galaxy S26 Ultra will be larger in length, width, and thickness, and likely weight. If you want a phone that’s easier to handle but still has a big screen, the Galaxy Plus is a good choice.  

Both phones will have at least IP68 water- and dust-resistance. Some high-end Chinese flagships now offer IP69 and IP69K ratings, which are better than IP68, but it’s not clear whether Samsung will do the same this year.  

The Galaxy S26 Ultra will include the built-in S Pen.  

There is no information yet about color options, but we expect a good selection, including some explosive colors, on samsung.com.  

Display Differences 

The Galaxy S26 Plus will have the same 6.7-inch Dynamic AMOLED 2X screen as last year, offering a refresh rate of up to 120 Hz, HDR support, and a peak brightness of over 2,600 nits.  

The Galaxy S26 Ultra will also feature a 6.9-inch Dynamic AMOLED 2X display with a 120 Hz refresh rate, HDR, and brightness close to 3000 nits.  

The prominent new display feature in the Galaxy S26 series is the privacy display. This uses AI software and Samsung’s Flex Window OLED panel to stop people next to you from seeing your screen while everything still looks normal to you. It’s similar to privacy screen protectors but built into the phone’s software.  

Both the Galaxy S26 Ultra and S26 Plus will come with Gorilla Glass Armor for strong scratch and drop resistance.  

Recent reports claim this feature won’t be exclusive to the Galaxy S26 Ultra, but will be supported by both the Galaxy S26+ and the Galaxy S26. However, since it requires the new OLED panel, it’s highly unlikely to arrive on older galaxies via a software update.  

Both phones will feature under-display fingerprint scanners, expected to be fast and accurate.  

Performance and Software 

Samsung is once again splitting its chip choices for the Galaxy S26 and S26 Ultra, and this time it might be a permanent change.  

In the USA, Canada, and China, the Galaxy S26 Plus will use the Qualcomm Snapdragon 8 Elite Gen5, the best high-profile 3nm chipset available for most Android makers.  

In other markets, the S26 Plus will use Samsung’s Exynos 2600 chip built on a 2nm process. Samsung says the chip is about five percent faster and 8% more efficient than the previous Exynos 2500.  

This doesn’t apply to the Galaxy S26 Ultra. The Ultra will use Qualcomm chips worldwide, with no Exynos version for the main flagship.  

It’s still unclear whether Samsung will offer 16GB of RAM in its flagships, as many Chinese brands do, or stick with 12GB. The latter seems more likely.  

Camera 

There are no major camera changes this year either.  

The Galaxy S26 Plus will have a 50MP main camera, likely with an upgraded sensor, a 12MP ultra-wide, and a 10MP telephoto with 3x optical zoom. It will retain its latest setup: A200MP main camera, A50MP 5x periscope, A50MP ultra-wide, and A10MP 3x telephoto. The change here could be either a larger sensor for the main camera (a 1/1.1-inch Sony sensor has been floated as a possibility) or a faster f/1.4 aperture.  

Both phones will keep their 12MP selfie cameras, but the lens might be wider to capture more background. This would be a nice upgrade.  

Battery Life and Charging 

Samsung might finally step away from the 5,000 mAh battery it has been using for the past five or six years on its Galaxy Ultra phones, but the change might not be as big as hoped. Samsung may move on from the 5,000 mAh battery it has used in Galaxy Ultra phones for years; however, the change may be small, with rumors pointing to a 5,200 mAh battery in the Galaxy S26 Ultra. The battery size is around 4,900 mAh, the same as last year’s S25 Plus.  

The Galaxy S26 Ultra will likely get a charging upgrade from 45 W to 60 W, enabling faster charging. The S26 Plus will likely keep its 45 W charging, which is now standard.  

There’s a high chance we’ll see full Qi2 implementations across the Galaxy S26 range, meaning both the Galaxy S26 Ultra and the Galaxy S26 Plus will be joining the party. This means wireless charging speeds potentially up to 25 W, but the bigger thing could be the use of Qi2 magnets in the rear. These could allow you to easily snap on compatible Qi2 Mag Safe or Pixel Snap accessories.  

Summary 

Samsung is playing it safe with its flagships again. There are no new product names, categories, or major changes in performance or design. It is a very familiar pair of devices that will retain their major selling points and key differentiators.  

The Galaxy S26 Ultra will likely be the top choice for power users, as it should be the most powerful Android phone in the US for some time. With the best camera and most features, it will be expensive but offer no compromises in early 2026. It will continue to be that weirdly positioned device that feels slightly off. It’s a flagship but not quite, and it’s compact but not as much as the Galaxy S26. It’s been nearly half a decade since the Galaxy S Plus phones, and they don’t have a valid reason for existence.

Source: Samsung Galaxy S26 Plus vs Galaxy S26 Ultra: Main differences to expect 

News Summary: 

• NVIDIA Bluefield 4 is the engine behind the NVIDIA Inference Context Memory Storage platform. A new AI-native storage system built for large-scale inference helps speed up and expand agentic AI.  

• This new storage processor is designed for agentic AI systems that need to handle long-term contexts, offering fast access to both long and short-term memory.  

• The Inference Context Memory Storage platform provides AI agents with long-term memory and enables fast cross-up sharing of context. This can increase tokens per second and power efficiency by up to five times.  

• With NVIDIA Spectrum-X Ethernet extended context memory, multi-term AI agents respond faster, boost throughput for each GPU, and make it easier to scale agentic inference.  

NVIDIA today announced that the NVIDIA Bluefield 4 data processor, part of the full-stack NVIDIA Bluefield platform, powers the NVIDIA Inference Context Memory Storage platform, a new class of AI-native storage infrastructure for the next frontier of AI.  

As AI models grow to trillions of parameters and employ multi-step reasoning, they generate vast amounts of context data. The data is stored in a key-value (KV) cache, which is important for accuracy, user experience, and continuity.  

Storing a KV cache on GPUs in the long term would slow down real-time inference in multi-agent systems. AI-native applications need a new scalable way to store and share this data.  

The NVIDIA Inference Context Memory Storage platform extends GPU memory, enables fast sharing across nodes, and can boost tokens per second and power efficiency by up to 5x compared with traditional storage.  

AI is changing the entire computing stack and now storage, said Jensen Huang, founder and CEO of NVIDIA. AI is no longer about one-shot chatbots, but intelligent collaborators that understand the physical world, reason over long horizons, and stay grounded in facts. AI uses tools to do real work and retain both short- and long-term memory. With Bluefield for NVIDIA and our software and hardware partners, we are reinventing the storage stack for the next frontier of AI.  

The NVIDIA Inference Context Memory Storage Platform increases KV cache capacity and speeds up context sharing across large AI system clusters. Persistent context for multi-turn AI agents also helps them respond faster, increases throughput, and encourages efficient scaling.  

The capabilities of NVIDIA Bluefield for Power Platform include:  

• NVIDIA Rubin offers cluster-level KV Cache capability, providing the scale and capability needed for long-context, multi-term agentic interface.  

• It delivers up to 5 times the power efficiency of traditional storage.  

• The platform uses the NVIDIA DOCA framework, along with the NVIDIA NIXL library and NVIDIA Dynamo software, to enable fast and smart sharing of KV cache across all nodes. This helps maximize tokens per second, reduce the time to the first token, and improve responsiveness in multi-turn tasks.  

• NVIDIA Bluefield 4 manages hardware-accelerated KV cache placement, removing metadata overhead, reducing data movement, and guaranteeing secure, isolated access from GPU nodes.  

• NVIDIA Spectrum-X Ethernet enables efficient data sharing and retrieval, acting as the high-performance network for RDMA-based access to AI-native KV cache.  

Companies like AIC, Cloudian, DDN, Dell Technologies, HPE, Hitachi Vantara, IBM, Notting, Pure Storage, Super Micro, Vast Data, and WEKA are some of the first to develop new AI storage platforms using Bluefield. These platforms are expected to be available in the second half of 2026.  

To find out more, watch N Media live at CES.

Source: NVIDIA BlueField-4 Powers New Class of AI-Native Storage Infrastructure for the Next Frontier of AI

Cognizant is expanding its partnership with Google Cloud to bring agentic AI to more enterprise operations at scale.  

The collaboration builds on a previous agreement to adopt Gemini Enterprise and aims to move from platform integration to immediate business implementation.  

At this new phase, Cognizant will use Microsoft 365 and Gemini Enterprise within its own operations. The goal is to boost productivity, improve employee experience, and make delivery more efficient.  

With Google Cloud, Cognizant helps companies turn their AI plans into managed, scalable systems that move from planning to applied use. Cognizant’s expanded partnership with Google marks a move from basic AI testing to the use of agentic AI on a much larger scale.  

This next step is designed to help companies go beyond simple automation and use AI systems that can autonomously manage complex multi-step business tasks.  

Procurement and supply chain leaders will find the focus on agentic solutions most important. These are AI systems built to work independently and achieve specific business goals.  

Cognizant will launch a new service for clients that brings together Gemini Enterprise and Microsoft 365. This is meant to help organizations move from manual work to AI-powered workflows, including tasks like creating content together and overseeing supplier communications.  

Google Cloud Global Ecosystem and Channels president Kevin H. Poranyi said: “Our partnership with Cognizant brings together advanced AI technology and deep industry expertise to help enterprises operationalize agentic AI.”  

Together, we are enabling organizations to deploy enterprise-ready AI solutions that deliver real business impact.   

To help standardize and scale its services, Cognizant is investing in the skills and tools needed to bring agentic AI to large businesses. It uses its agent development lifecycle (ADLC) to incorporate AI at every step, from design to rollout.  

Cognizant Ignition, powered by Gemini, is designed to accelerate delivery, discovery, and prototyping, and strengthen clients’ data systems.  

With Cognizant Agent Foundry, the company provides no-code tools and ready-made models for things like AI-powered contact centers and advanced order management. Clients can access these through Google Experience Zones and GenAI Studios, both of which are supported by a team of trained Gemini technology experts.  

Cognizant sees this as a model for companies to use agentic AI at scale, moving from choosing platforms to having systems ready to run. The partnership with Google Cloud aims to give organizations ways to manage their AI and measure results.  

Cognizant Core Technologies Plus Insights President Annadurai Elango said the partnership reinforces Cognizant’s position as an AI builder, a new kind of services partner focused on creating purpose-built, enterprise-grade solutions that deliver real business outcomes.  

Cognizant brings together the optimal combination of people and technology, including proprietary IP and deep service expertise, to build industry-specific platforms, embed context into systems, and co-create agentic solutions customized to each client’s businesses.  

The expanded partnership was first announced in October 2025, when Cognizant shared plans to help clients automate using Gemini Enterprise across services like NeuroAI and Agent Foundry. Both companies used Gemini Enterprise, Vertex AI, and the customer engagement suite to deliver results across different industries.  

Gemini Enterprise lets employees use Google’s latest AI models and offers a unified experience for managing AI agents in one place.  

Alongside its tech updates, Cognizant also shared its fourth-quarter and full-year 2025 fiscal results. Net income was $648 million in Q4 2025, up 18.6% from the same time in 2024.  

Revenue for the 4th quarter was $5.3 billion, up 4.9% year-on-year or 3.8% in constant currency terms over the full year. Revenue reached $21.1 billion, a 7% increase or 6.4% in constant currency terms.  

Cognizant expects first-quarter 2026 revenue to be between $5.36 billion and $5.44 billion, representing growth of 4.8% to 6.3% or 2.7% to 4.2% in constant currency.  

For all of 2026, Cognizant estimates revenue between $22.14 billion and $22.66 billion, an increase of 4.9 to 7.4% or 4 to 6.5% when adjusted for currency changes.  

In late January, Cognizant partnered with Cognizant to use AI in software engineering. This will bring Cognizant’s DevIn AI, an autonomous software engineer that can handle development tasks independently, into business settings.

Source: Cognizant scales agentic AI operations through Google Cloud alliance

Today, OpenAI is launching a research preview of GPT-5.3 Codex Spark, a smaller version of GPT-5.3 Codex and its first model built for live coding. Codex Spark is the first result of our partnership with Cerebras, announced in January. It’s designed to feel almost instant on ultra-low-latency hardware, delivering over 1,000 tokens per second while staying highly effective for real-life programming tasks.  

We are sharing Codex Spark on Cerebras as a research preview for ChatGPT Pro users, so developers can start experimenting early. At the same time, we work with Cerebras to ramp up data center capacity, harden the end-to-end user experience, and deploy our larger frontier models.  

Our latest models are especially good at handling long-running tasks, working on their own for hours, days, or even weeks. Codex Spark is our first model built for instant use with Codex, so you can make targeted edits, adjust logic, or refine interfaces and see results right away. Now Codex supports both big, ongoing projects and quick, in-the-moment work.  

We look forward to learning from developers and using your feedback as we expand access.  

At launch, Codex Spark has a 128K context window and is text only. During the research preview, Codex Spark will have its own rate limits, and usage will not count towards standard rate limits. However, when demand is high, you may see limited access or temporary queuing as we balance service reliability across users.  

Speed and Intelligence 

Codex Spark is built for interactive work where speed is just as important as intelligence. You can work with the model in real time, interrupt or redirect it as needed, and quickly try out new ideas with fast responses. Since it’s tuned for speed, Codex Spark keeps things simple by making only minimal targeted edits and running tests only when you ask.  

Coding 

Codex Spark is a powerful small model designed for fast results on SWE Bench Pro and Terminal Bench 2.0, which tests software engineering skills. GPT-5.3 Codex Spark performs well and completes tasks much faster than GPT-5.3.  

Latency Improvements For All Models 

As we trained Codex Spark, it became apparent that model speed was just part of the equation for instant collaboration. We also needed to decrease latency across the full request-response pipeline.  
 
We implemented end-to-end latency improvements in our harness, benefiting all models. We streamlined how responses stream from client to server and back, remote key pieces of our inference stack, and reworked how sessions are initialized so that the first viable token appears sooner, and Codex stays responsive as you iterate. By introducing a persistent WebSocket connection and targeted optimizations in the responses API, we reduced:  

• per-client/server round-trip overhead by 80%  

• per-token overhead by 30%  

• time to first token by 50%  

The WebSocket path is enabled for Codex Spark by default and will become the default for all models soon.  

Powered by Cerebras 

Codex Spark runs on the Cerebras Wafer Scale Engine 3, a specialized AI accelerator built for high-speed inference, providing Codex with a low-latency serving option. We worked with Cerebras to add this fast path to our main production system. Codex Spark works smoothly with the rest of our models and prepares us to support future ones.  

GPUs remain the main component of our training and inference systems and offer the most cost-effective tokens for general use. Cerebras adds to this by handling tasks that require very low latency, making Codex feel more responsive as you work. You can also combine GPUs and Cerebras for the best performance on a single workload.  

Availability and Details 

Codex Spark is rolling out today as a research preview for ChatGPT Pro users in the latest versions of the Codex app, CLI, and VS Code extension. Since it uses special low-latency hardware, it has its own rate limit that may change based on demand during the preview. We are also offering Codex Spark in the API to a small group of design partners to learn how developers want to use it in their products. We will expand access further in the coming weeks as we continue improving our integration and learn more with the developer community about where fast models shine for coding. We will introduce even more capabilities, including larger models, longer context lines, and multi-modal input.  

Codex Spark has the same safety training as our main models, including cybersecurity training. We reviewed Codex Spark as part of our usual deployment process, which includes access to cyber and other capabilities. We found that it does not have a realistic chance of meeting our preparedness framework threshold for high capability in cybersecurity or biology.  

What’s Next? 

Codex Spark is the initial step toward a codex that offers two modes:  

1. Long-term Reasoning and Execution  

2. Instant Collaboration for quick changes  

Over time, these modes will blend. Codex will let you stay in a close interactive loop while sending longer tasks to sub-agents in the background or spreading tasks across many models at once when you need speed and coverage, so you won’t have to pick up just one mode at the start.  

As models improve, interaction speed becomes a greater challenge. Ultra-fast inference helps close the gap, making Codex easier to use and opening new possibilities for anyone turning ideas into working software.

Source: Introducing GPT‑5.3‑Codex‑Spark

Apple has announced an event for March 4, where we could see a new budget iPhone, updated iPads, and maybe even a more affordable MacBook.  

Please join us in person for a special Apple experience in New York. The invitation needs to be next to the Apple logo in yellow, green, and blue. The event begins at 9 am ET, but Apple hasn’t said if there will be a live stream or shared any extra hints. PCMag will cover the event and share all details. In the meantime, here’s what we expect from Cupertino.  

iPhone 17E 

The iPhone 17E is likely to be one of the new devices. Rumors say it will keep the iPhone 16E’s design but get some upgrades inside. These include:  

• The A19 chip from the iPhone 17  

• Apple’s C1X modem for 5G and LG E  

• The N1 chip for WiFi 7 and Bluetooth 6  

Apple says the C1X is twice as fast as the C1 and N1 chips, improving the overall efficiency and reliability of features like Personal Hotspot and AirDrop. iPhones with the C1X (and C1) and iOS 26.3 also support Limit Precise Location, a setting that lets you control how much location data your iPhone shares with your cellular network.  

In PCMag’s review, we found that the iPhone 16e gracefully fills the role as the most affordable member of Apple’s iPhone family without damaging the core experience. It starts at $599 for the 128 GB, which is $100 less than the base iPhone 16 and $200 less than the standard iPhone 17. We’ll have to see if Apple can afford that $599 price for the 17E, or if the memory crunch will affect Cupertino’s lineup.  

The 17e is likely the only new iPhone to be announced on March 4. The iPhone 16e also launched in late February last year. Rumors last year suggested we’ll see four smartphones at Apple’s usual September event. The iPhone 18 Pro, plus the second-generation iPhone Air, and a foldable, the standard iPhone 18, and the 18e are not expected until early 2027.  

No matter what Apple announces, the event is set for one week after Samsung’s next unpacked event on February 25, where Samsung will reveal the new Galaxy S26 series.  

Apple Intelligence for the 12th Gen iPad 

If you are looking to update your iPad, Apple is expected to refresh the iPad Air lineup with an M4 chip, while the entry-level iPad will get an A18 chip. The current 11- and 13-inch iPad Airs use the M3, and last year’s 11th-gen iPad has an A16. Upgrading to the A18 means the iPad will support Apple intelligence, which the iPad Airs already have thanks to their M series chips.  

Low-cost MacBook and new MacBook Pros 

Rumors have been swirling about a possible cheap MacBook. The 12-inch device could run an Apple A-series processor, usually found in iPhones and iPads, rather than the more powerful and expensive M-series. Bloomberg’s Mark Gurman says it will be the Apple 18 Pro found in the iPhone 16 Pros. He also predicts some new color options.  

In October, we explored how a budget A19 MacBook might perform and found it would fit nicely among today’s budget laptops. Is that why the M1 MacBook Air is MIA at Walmart? Earlier this month, Gurman also mentioned a MacBook Pro with the M5 Pro and M5 Max chips. Signs that a launch is coming might soon include fewer units in stock and shipping delays for the current top-end MacBook Pros.

Source: New iPhone, iPads, and Macs? Apple Announces March 4 Event 

Samsung Electronics announced it will showcase its connected AI-powered kitchen vision at the Kitchen & Bath Industry Show (KBIS) in Orlando, Florida, from February 17 to November 18 by combining its bespoke AI lineup with Dacor, its premium built-in kitchen brand. Samsung is introducing a new wine culture that reinvents contemporary life with smart storage, elegant design, and AI innovation. This sets a new standard for smart and reliable homes.  

“The kitchen is becoming the center of home life, and Samsung keeps introducing smart solutions that connect appliances, services, and daily experiences,” said Sang Jik Lee, Executive Vice President and Head of Sales and Marketing for the Digital Appliances Business at Samsung Electronics. At KPIS, we are showing how Dacor redefines modern living by combining function, style, and smart technology, molding the future of the kitchen where every detail enhances both design and utility.  

Dacor’s Premium Wine Culture and Innovative Design of the Hidden Kitchen 

At the exhibition, a special display highlights Dacor’s products designed for wine lovers. Featured items include:  

• The 24-inch under-counter wine column  

• Built-in wine dispenser  

The dispenser holds up to four bottles and has dual temperature zones to keep red and white wines at their ideal temperatures with a simple touch. Users can choose the right amount of wine.  

Dacor is also presenting a sleek, organized kitchen design that merges with the home’s look when not in use. These hidden kitchen concepts are built into curved columns or furniture and include a full line-up of refrigerator, cooktop, and the 2025 Luxe Red Design award-winning 24-inch dishwasher. This approach delivers both usefulness and style.  

Dacor’s premium appliances also work with SmartThings, letting users manage key functions through their connected devices.  

Bespoke AI: Smart, Connected, And Reliable Kitchen Innovation 

The Bespoke AI Refrigerator Family Hub is also on display at KBIS, featuring the new AI Vision technology with advanced camera-based recognition. Users can easily check what’s inside their fridge and manage ingredient lists. Previously, the system could recognize up to 37 types of fresh food and 50 types of pre-registered processed foods directly on the device. Samsung plans to improve this feature to recognize even more food items.  

Visitors at KBIS can also see a variety of Bespoke AI products, including dishwashers, ovens, and laundry appliances. These products stand out because they are smart things that let users easily monitor them from their smartphones. Bixby for voice control adds to the experience. For example, a user can ask the refrigerator, “Hi Bixby, can you recommend a recipe based on my food list?” It will provide recipes and step-by-step cooking instructions.  

Since reliability is a major factor in building a strong, stable, connected ecosystem, Samsung has dedicated a section of its exhibit to its commitment to ensuring the quality of Bespoke AI products and services available to consumers after purchase, as well as to routine software updates. Samsung offers Home Appliance Remote Management (HARM), a service that helps consumers understand and manage appliance status through monitoring, diagnosis, and remote service support.  

Samsung at KBIS 2026 

KBIS takes place from Feb 17-19. Samsung’s exhibit is at booth W2073 in the West Hall of the Orange County Convention Center.

Source: 

Samsung Highlights Bespoke AI and Dacor Kitchen Appliance Innovations at KBIS 2026

Google has added Agentic Vision 3 to its Gemini 3 Flash model to help it better understand images and make fewer mistakes in visual tasks.  

Agentic Vision lets the model work like an active investigator. It follows a think-act-observe process to examine and modify images by running code.  

This update helps prevent the AI from making guesses when image details are small or hard to see.  

Main features of Agentic Vision 

• Think, Act, Observe loop: The model first looks at the question and image (think). Next, it writes Python code to change or study the image (like cropping or adding nodes) (act). Finally, it checks the updated image for greater context before answering (observe)  

• 5-10% quality boost: running code with Agentic Vision in Gemini 3 Flash makes the model 5-10% more accurate on most visual tests.  

• Visual scratch pad: the model can add notes or marks directly on images, so its analysis uses actual image pixels.  

• Reduced hallucination: Agentic Vision uses Python code to perform tasks such as counting small objects, reading distant text, or studying tables. This stops the model from making random guesses that lead to mistakes.  

Main Uses 

• Zooming and inspection: The model can zoom in on small or blurry details on its own.  

• Visual Math and Plotting: Agentic Vision pulls data from tables in images, does the math, and makes charts rather than guessing the numbers.  

• Interactive annotations can draw boxes and labels to count items in busy images accurately.  

Where to Find it 

You can find Agentic Vision in:  

• Google AI Studio: Developers can turn on code execution under “Tools” in the playground.  

• Vertex AI: Available through the Gemini API.  

• Gemini app: added under the Thinking Model option.  

Future updates will make these features automatic and bring them to other Gemini models.  

Frontier AI models like Gemini usually process the world in a single static glance. If they miss a small detail, such as a microchip’s serial number or a distant street sign, they have to guess.  

Agentic Vision in Gemini 3 Flash changes image understanding from a static process to an active one. It treats vision as an investigation by combining visual reasoning with code execution. The model can plan to zoom in, inspect, and manipulate images step by step, grounding its answers in visual evidence.  

Allowing code execution with Gemini 3 flash gives a steady 5-10% quality boost on most vision benchmarks.  

Agentic Vision: A New Frontier in AI Capability 

Agentic Vision brings a think-act-observe loop to image understanding tasks.  

1. Think: the model examines the user’s question and the initial image, then generates a step-by-step plan.  

2. Act: The model writes and runs Python code to work with images, such as cropping, rotating, and adding nodes. It also analyzes images by running calculations or counting objects.  

3. Observe: The changed image is added to the model’s context. This helps the model review the new data with more context before giving a final answer.  

Agentic Vision in Action 

When you enable code execution in the API, you open up a range of new possibilities. Our demo app in Google AI Studios shows many of these in action. Developers from large companies using the Gemini app to small startups are already using this feature for a variety of use cases, such as:  

1. Zooming and Inspecting 

Gemini 3 Flash automatically zooms in on small, detailed features. Planchecksolver.com, an AI tool for checking building plans, increased its accuracy by 5% after enabling code execution with Gemini 3 Flash. This allowed the platform to inspect high-definition images in a step-by-step fashion. In a video of the backend logs, you can see Gemini 3 Flash generate Python code to crop and analyze specific areas, such as roof edges or building sections, into new images. By adding these cropped images back into its context, the model can visually check its reasoning and confirm that plans meet complex building codes.  

2. Image Annotation 

In the Agentic Vision, the model can interact with its environment by adding notes or drawings to images rather than only describing what it sees. Gemini 3 Flash can run code or draw directly on the image, helping to show its reasoning.  

In the example below, the model is asked to count the fingers on a hand. In the Gemini app, to avoid mistakes, it uses Python to draw boxes and numbers over each finger it finds. This visual scratch pad helps ensure the answer is accurate down to the pixel.  

3. Visual Math and Plotting 

Agentic Vision can read complex tables and use Python code to create visualizations of the results.  

Standard language models can make mistakes when doing multi-step visual math. Gemini 3 Flash avoids this by using a reliable Python environment for calculations. In the example below, from our demo app in Google AI Studio, the model finds the raw data, writes code, sets the previous SOTA to 1.0, and creates a matplotlib bar chart. This way, the results are based on real execution, not guesses.  

What’s Next? 

We are only at the beginning with Agentic Vision.  

• More implicit code-driven behaviors: Right now, Gemini 3 flash is great, automatically zooming in on small details. Other features, like rotating images or doing visual math, still need a clear prompt to work. We are working to make these actions automatic in future updates.  

• More Tools: We are also exploring ways to give Gemini models more tools, such as web search and reverse image search, to help them better understand the world.  

• More model sizes: We also plan to bring this feature to more of our models, not just Flash.

Source: Introducing Agentic Vision in Gemini 3 Flash