Source: Microsoft

Microsoft has announced a major consolidation of its AI lineup by merging the consumer Copilot app with Microsoft 365 Copilot into a single platform. The updated super app brings together chat, image generation, and direct connection to Word, Excel, and Outlook while strictly isolating personal and work data. Core features will remain free, with higher capacity reserved for Microsoft 365 subscribers. The transition also marks the retirement of legacy features like Group Chat, Deep Research, and Podcasts.

Microsoft plans to unify its Copilot applications

Microsoft is planning to merge its various individual Copilot offerings into a single experience. This integration will combine the consumer Copilot application with the Microsoft 365 Copilot app. Previously, the tech giant operated three distinct Copilot platforms aimed at different user groups. Microsoft 365 Copilot was tailored for corporate and enterprise clients, GitHub Copilot catered to developers, and the standard Microsoft Copilot app served general consumers. Each product maintained a dedicated app and distinct web interface.

This strategic pivot aligns with wider shifts across the artificial intelligence sector. OpenAI recently consolidated ChatGPT, ChatGPT Work, and Codex into a unified ChatGPT platform. Following this direction, Microsoft aims to streamline its Copilot suite into a single offering serving multiple user segments. During the latest earnings call, Microsoft CEO Satya Nadella highlighted this strategy, stating that the company will “bring chat, Cowork, Autopilots, and Code together into one flagship super app spanning both consumer and commercial experiences.”

The combined Copilot super app is scheduled for release in the July-September quarter. The platform will support connections to third-party software and enterprise applications, including Agent 365, CRM, and ERP systems, using skills and plug-ins. Furthermore, Microsoft will implement a business model based on per-user licensing alongside consumption-based pricing.

Support documentation outlines the upcoming unified interface

Microsoft has released an official support document detailing how the consumer Copilot app and the Microsoft 365 Copilot app will merge. Interestingly, the support article omits any mention of software development or coding tools, despite Satya Nadella’s previous remarks during the earnings presentation.

According to the document, the updated software will combine core Copilot chat and image generation capabilities with embedded access to Microsoft 365 applications such as Word, Excel, and Outlook. Users will also be able to link their email accounts, calendars, and cloud storage to supply the AI assistant with broader operational context. The official statement summarizes the planned transition: 

We’re updating Copilot to create a simpler, more cohesive experience for everyone. Depending on the account and device you use, you will see changes to the Copilot app including changes to appearance and functionality, such as navigation, feature availability, or sign-in experience.

Users will be able to log in using personal, work, or school credentials, with support for switching between multiple account types within the same interface. Enterprise data controls and personal privacy boundaries will remain isolated. As noted in the documentation:

 Work and personal accounts remain separate. Your organization’s security, privacy, compliance, and administrative controls continue to apply when you use Copilot with your work or school account.

Web users will be automatically redirected to the updated portal, while mobile users may need to manually update or reinstall the application. For individuals currently utilizing both apps with a personal account, chat histories and generated content will be merged into the unified platform.

Discontinuation of consumer features ahead of the merge

The unification comes alongside the sunsetting of several consumer-facing capabilities. In July, Microsoft revealed that it would discontinue the Deep Research and Podcasts tools within the consumer Copilot app starting August 18, 2026. The company has now confirmed that Group Chat will also be retired on the same date. Existing message histories, chat threads, and generated images from Group Chat will not migrate to the updated platform.

Microsoft advises users to manually save any desired messages into a document or notes app and download any created images before their accounts are updated. Regarding audio content, the company clarifies that customers will lose the ability to create or access Podcasts in Copilot once the feature is retired, though they can save individual episodes beforehand by using the download option in the podcast menu. Additionally, Microsoft cautions that users might encounter temporary feature gaps or unavailable tools while the updated version of Copilot is being rolled out.

Pricing tier details and data management policies

The baseline features of the consolidated Copilot app will remain free to use. Standard capabilities – including conversation, image generation, and file uploads – will continue without charge, subject to system capacity limits. However, subscribers to Microsoft 365 plans will receive higher usage allowances, access to advanced AI models, custom agent support, and tools for multi-step tasks. The official document notes:

You can continue to chat with Copilot, create images, upload files and more for free, subject to available capacity and limits. When you hit a limit, you can choose to upgrade to a paid option or come back later.

Data storage and privacy architecture will also see adjustments. Documents and images shared or generated within the standalone Copilot app will transfer directly to OneDrive, accessible via mobile, desktop, or web clients.

On the privacy front, Microsoft confirms that commercial boundaries remain intact, explaining that data entered into the work experience via Microsoft Entra does not flow into the personal Microsoft account experience, and vice versa.

Microsoft advances AI image generation with release of MAI-Image-2.6

Source: Microsoft

Microsoft has unveiled its newest image generation model, MAI-Image-2.6. The launch propels the company past major competitors, including Meta, Google, ByteDance, and xAI, placing it #2 on the Arena leaderboard behind only OpenAI’s GPT-Image-2. The updated model scored +79 Elo points higher than MAI-Image-2.5, delivering significant advancements in text rendering, portraits, 3D graphics, and commercial visuals. 

MAI-Image-2.6 climbs to second place on the Arena leaderboard 

Microsoft launched its first in-house text-to-image model, MAI-Image-1, in October 2025, entering the Arena leaderboard at #9. In March 2026, the company introduced MAI-Image-2, a second-generation upgrade that delivered substantial improvements in image quality and jumped to #3, ranked just behind Google’s gemini-3.1-flash-image-preview and OpenAI’s gpt-image-1.5-high-fidelity.

In May, Microsoft released MAI-Image-2.5, which held the #3 position on the leaderboard. Last month brought the launch of MAI-Image-2.5-Pro, a high-fidelity model aimed at detailed image generation, editing tasks, and precise inside-image text rendering.

Source: Arena.AI

On August 11, Microsoft has introduced its latest text-to-image model, MAI-Image-2.6. This release puts the company ahead of rival image generation models from Meta, Google, ByteDance, and xAI. On the Arena leaderboard, the new model secured the #2 spot, trailing only OpenAI’s GPT-Image-2. Compared to its predecessor, MAI-Image-2.5, the 2.6 release scored an additional +79 Elo points in the text-to-image category.

Technical capabilities and commercial focus

The new MAI-Image-2.6 delivers improvements in text rendering, portraits, and 3D imagery. Microsoft designed the model to produce polished commercial and photorealistic visuals suitable for product design, branding, and cinematic projects.

The update simplifies workflows that require multiple reference inputs and provides better grounding. Additionally, it gives users greater control over reasoning, format selection, and resolution output.

With every MAI-Image release, we’re focused on moving the quality frontier forward. MAI-Image-1 gave us the foundation. MAI-Image-2 made a major jump in photorealism, text and creative range. MAI-Image-2.5 pushed further into professional-grade imagery and editing. MAI-Image-2.6 continues that climb with broad gains across different categories that we know our users care about most – the company stated.

Users can test MAI-Image-2.6 for text-to-image generation on Arena. Microsoft began rolling out the model to MAI Playground and Microsoft Foundry, before expanding availability across other company products and services. Additional technical details regarding the release will be shared in the coming weeks.

OpenAI speeds up GPT-5.6 Sol without losing intelligence

Source: OpenAI

OpenAI officially announced the GPT-5.6 Sol Ultrafast mode in its API. Powered by Cerebras’ Wafer-Scale Engine architecture, the new tier generates up to 750 tokens per second – running up to 14 times faster than Standard processing without sacrificing model capability. This speed boost is designed for time-critical tasks like incident response, financial research, and real-time support. The Ultrafast mode is currently rolling out in a limited preview to select customers as capacity grows.

OpenAI launches GPT-5.6 Sol Ultrafast mode powered by Cerebras

OpenAI has announced GPT-5.6 Sol Ultrafast mode, a new service tier in the OpenAI API that delivers up to 750 output tokens per second. Powered by Cerebras’ Wafer-Scale Engine architecture, which uses 44GB of on-chip SRAM to eliminate GPU memory bottlenecks, this mode runs OpenAI’s most capable model up to 14 times faster than Standard processing without reducing intelligence.

Benchmarking from Artificial Analysis shows Ultrafast is 11 times faster than Claude Fable 5 and 5 times faster than Claude Opus 4.8 in Fast mode. In the Humanity’s Last Exam test, Sol Ultrafast processed 2,500 questions in 11 hours and 11 minutes compared to 78 hours and 27 minutes for Fable 5, maintaining similar accuracy.

High speed enables real-time enterprise use cases

The speed increase allows frontier AI to be integrated directly into time-critical workflows:

• Incident response: Analyzing logs and code changes during active system outages to speed up root-cause analysis and fixes.
• Finance and security: Assessing transactions and market signals dynamically as conditions shift.
• Customer support and voice: Resolving complex, multi-step issues mid-conversation without pauses.
• Commerce: Handling product questions, recommendations, and checkout obstacles instantly.
• Research and development: Turning overnight experiment runs into rapid, real-time feedback loops during the workday.

Limited preview access and rollout plans

OpenAI developers are already using Sol Ultrafast internally for incident analysis and iterative research. Due to limited initial Cerebras infrastructure capacity, the Ultrafast mode is currently available only in a restricted preview to select customers, with broader access planned as capacity expands.

OpenAI expands Daybreak program with specialized GPT-5.6 cybersecurity model

Source: OpenAI

OpenAI has expanded its Daybreak cybersecurity program and introduced GPT-5.6-Cyber, a model built for advanced security research. As AI capabilities accelerate, the company aims to put frontier tools into the hands of defenders before threat actors can exploit software vulnerabilities autonomously at scale.

Initially launched in June alongside GPT-5.5-Cyber, the Daybreak program now features two access tiers: Blue and Red. Daybreak Blue provides defenders with general-purpose frontier models like GPT-5.6 Sol under relaxed guardrails suitable for patch validation, incident response, and secure code reviews. Meanwhile, Daybreak Red grants access to specialized models like GPT-5.6-Cyber, designed for dual-use cybersecurity tasks such as finding zero-day vulnerabilities and developing exploit chains.

Higher request completion across specialized benchmarks

Built on top of GPT-5.6 Sol, GPT-5.6-Cyber is specifically trained to reduce refusals on legitimate defensive tasks that standard models often block. On OpenAI’s internal Advanced Cybersecurity Completion Rate evaluation – which tests scenarios like privilege escalation and authentication bypasses – GPT-5.6-Cyber achieved a 95% completion rate. By comparison, GPT-5.5-Cyber reached 57.3%, while standard GPT-5.6 Sol completed only 1.5% with standard safeguards and 2% via Daybreak Blue.

Beyond handling complex requests, the new model demonstrates improved performance across specialized technical benchmarks. On ExploitGym, GPT-5.6-Cyber outperformed earlier models in converting known bugs into working exploits within controlled environments. It also led in zero-day discovery evaluations, though GPT-5.6 Sol outperformed it in certain report-writing tests due to generating more detailed documentation.

Real-world discoveries backed by mandatory hardware keys

In real-world testing, GPT-5.6-Cyber successfully identified high-severity bugs across several major platforms. Most notably, it discovered two novel vulnerabilities in Chrome’s V8 JavaScript engine (CVE-2026-15903) that could be chained to corrupt memory and escape the V8 heap sandbox. Google has since patched the flaw. The model also identified over 400 privilege escalation vulnerabilities in an operating system kernel, three critical issues in a popular database, and five mobile operating system flaws.

Source: OpenAI

To mitigate misuse risks associated with reduced safeguards, OpenAI enforces identity verification, account security monitoring, legal attestations, and approved-use restrictions. Starting September 1, 2026, individual accounts in Daybreak will require hardware security keys. OpenAI also recommends using Codex auto-review modes, isolating security workflows in controlled sandboxes, and establishing clear operational scopes.

Microsoft reveals upgraded MAI-Code-1.1-Flash model

Source: Microsoft

Microsoft has launchedv an upgraded version of its coding model, designed to regain ground against rising market rivals. The new variant brings notable speed and performance improvements to everyday development tasks while cutting costs to a fraction of the previous rate. Featuring expanded vision capabilities to analyze visual inputs directly, the model is already rolling out across supported developer environments, paving the way for the eventual retirement of its predecessor. 

Responding to competition from market rivals

Back in June, Microsoft first revealed MAI-Code-1, an inference-efficient coding model tuned for GitHub Copilot workloads. The Flash variant of this solution was made available in GitHub Copilot as an alternative to models created by OpenAI and Anthropic.

While the pricing and efficiency of MAI-Code-1-Flash looked competitive at the time of its announcement, it was quickly overshadowed by affordable and powerful Chinese models such as GLM-5.2 and Kimi K3, as well as OpenAI’s GPT-5.6 Luna. In response to these challenges, Microsoft has announced MAI-Code-1.1-Flash, an upgraded version that delivers better coding performance while significantly reducing cost and token usage.

Higher performance in key benchmarks

According to Microsoft, the new model performs 22% better on Terminal-Bench 2.1 when used through the GitHub Copilot CLI and delivers a 15% improvement on .NET tasks.

We learned from developer feedback that CLI tasks and .NET performance mattered, so that’s where we focused. The result: a 22% improvement on Terminal-Bench 2.1 in GitHub Copilot CLI and a 15% improvement on .NET tasks – explains Microsoft.

Improvements also covered overall code quality and operational speed. According to the creators, tokens stream 25% faster while the model uses 25% fewer tokens to complete the same task.

MAI-Code-1.1-Flash produces higher quality code, at 25% greater token efficiency, and at a quarter of the cost compared to the model we launched in June at Microsoft Build. This small, efficient, coding workhorse is now in production in GitHub Copilot – the company stated.

The developers note that the most critical evaluations occur during daily usage.

Benchmarks are useful guides but production is where the rubber meets the road. Most importantly, code survival rose 4% and return visits increased 9% – added Microsoft.

Thanks to higher efficiency, developers can count on faster output and more effective resource usage.

Dramatic reduction in token pricing

Improvements to training and serving efficiency have allowed Microsoft to offer MAI-Code-1.1-Flash at one-quarter of the price of the original model.

Better training and serving efficiency let us offer a stronger, faster model at one quarter of the price of 1.0 – and pass those savings reliably to customers. We achieved this by optimizing for real-world use across more than hundreds of thousands of reinforcement-learning environments in GitHub Copilot – reported Microsoft.

As per the GitHub Copilot pricing table, MAI-Code-1.1-Flash costs $0.20 per 1 million input tokens, $0.02 per 1 million cached input tokens, and $1.20 per 1 million output tokens. These rates are dramatically lower than MAI-Code-1-Flash, which cost $0.75 / $0.075 / $4.50 per 1 million tokens. For annual Copilot subscribers, the model is charged at a 0.25x premium request multiplier.

The process of continuous development relies on ongoing learning from experience.

Yhe loop is simple: ship, learn, improve, repeat. That’s the MAI hill climbing machine – summarized Microsoft.

New vision capabilities, retirement of previous variant

Beyond coding improvements, MAI-Code-1.1-Flash adds native vision support for understanding images.

MAI-Code-1.1-Flash is currently rolling out across GitHub Copilot, including VS Code, Visual Studio, JetBrains IDEs, Copilot CLI, GitHub Mobile, and other supported experiences. Free and Student users can access it through automatic model selection, while paid users can manually select it.

With the upgraded model now available, GitHub has scheduled the retirement of MAI-Code-1-Flash across all Copilot experiences for September 10, 2026.

Gemini app surpasses one billion monthly active users

Source: joshwoodward, x.com

Google’s Gemini app has officially crossed the one billion monthly active users threshold, marking it as the fastest-growing product in the tech giant’s history. Launched in February 2024, the consumer application joins 13 other Google services that boast over a billion users. The rapid expansion saw user counts rise from 400 million at Google I/O 2025 to 750 million earlier this year, reaching 900 million in May and 950 million in late July before breaking the record this week. 

Ecosystem integration drives rapid expansion

The surge in adoption stems largely from Google’s distribution strategy across its existing platforms, including Android, Chrome, and Search. According to Gemini chief Josh Woodward, the one-billion figure accounts strictly for direct app interactions across web, Android, iOS, and Chrome, excluding model usage integrated into tools like Search or Gmail.

Replacing Google Assistant with Gemini on Android allowed hundreds of millions of users to access the platform via standard shortcuts without downloading a new application.

The service is also securing a strong foothold on competitor hardware, recording over 100 million active iOS users. Engagement metrics highlight diverse interaction habits: around 63% of users interact via voice, 20% of Gemini Live sessions utilize camera or screen sharing, 38% of school-related queries contain attachments, and daily image generation exceeds 150 million.

Competition intensifies in consumer AI

While OpenAI recently reported over one billion weekly active users for ChatGPT, the metrics differ, as ChatGPT maintains a higher usage frequency on a billion-user scale. Nevertheless, the parallel achievements setup a competitive dynamic between OpenAI’s established consumer mindshare and Google’s hardware and ecosystem reach. As both platforms scale past massive user thresholds, the battle for consumer dominance will depend on how effectively each company leverages its unique strengths.