AI Pulse & Updates

Qwen3.8-27B Open-Weight Model Release

Alibaba released its large language model "Qwen3.8-27B" as an open-weight model at midnight on August 15, distilling technology from the higher-tier Qwen3.8 Max into a lighter form capable of running on high-end graphics cards.

V4-Pro General Availability Release With Agent Capabilities

DeepSeek released its V4-Pro model designation DeepSeek-V4-Pro-0813 with focus on agent capabilities, marking the official general availability after preview period.

University of Toronto partners with Cohere on responsible AI adoption

University of Toronto announced a multi-year partnership with Cohere to advance responsible AI adoption at scale and take ownership of its AI future.

Qwen3.8-27B model scheduled for release

Qwen3.8-27B is a renewal of the beloved Qwen model scheduled for release on August 14, 2026, with 4,768 users awaiting the release.

DeepSeek V4-Pro officially launches with agent capabilities

DeepSeek released V4-Pro model on its app, web, and API, focusing on agent capabilities with tasks where AI systems use tools and complete multi-step workflows.

Claude adds smarter Slack collaboration with Claude Tag

Claude now uses full channel context and memory to participate proactively in Slack conversations, improving contextual awareness by 30%.

Previewing Ultrafast mode: GPT-5.6 Sol at up to 14X the speed

OpenAI announced GPT-5.6 Sol with Ultrafast mode delivering significantly faster performance, advancing capabilities in applied AI.

DeepSeek V4 Pro 0813 reaches general availability

DeepSeek V4 Pro 0813 is the general-availability release of DeepSeek's 1.6T-parameter MoE flagship, live on August 12, 2026.

In talks to buy artificial intelligence startup Decart AI for about $6 billion

Anthropic is in talks to acquire Decart AI for approximately $6 billion, marking its largest known acquisition ahead of an anticipated IPO.

Appoints Koray Kavukcuoglu as New Head of DeepMind

Koray Kavukcuoglu is becoming head of DeepMind, Google's AI unit, and will report directly to Google CEO Sundar Pichai, overseeing Gemini model development, frontier AI research and the Gemini app and developer teams, signaling a leadership shift.

Claude Code Fixes Session, Sync, and Tooling Issues

Claude Code fixes a broad set of session, sync, and tooling issues, including redrawing failures, Git detection on Windows, Remote Control resume leaks, self-hosted runner reliability, and memory cleanup, improving developer experience.

Launches Public Beta of Grok Bot AI Agent

xAI launched the public beta of Grok Bot on August 11, 2026 — a tool that started as an internal prototype and, according to the company, has already changed how the xAI team works day to day, with a Grok 4.6 release coming later in the week.

Launches Open Platform for Qwen, Enabling Third-Party Developers to Build Custom AI Agents

Alibaba launched an open platform for its AI model Qwen, allowing third-party developers to build custom AI agents across mobile phones, personal computers and smart glasses, expanding the Qwen ecosystem's developer reach.

OpenAI and APA Advance Responsible AI for Youth

OpenAI published a security announcement on August 10, 2026 on advancing responsible AI for youth with the APA.

Releases Muse Glimmer Open-Source AI Model

Meta released Muse Glimmer with a "permissive" open source license for a lightweight AI model that can run on a personal computer, part of an announcement as CEO Mark Zuckerberg warned of risks if control of advanced AI is concentrated under a select

Strikes $9.1 Billion Deal with Riot Platforms

Anthropic secured a $9.1 billion computing deal with Riot Platforms, a Bitcoin mining company, underscoring efforts to secure enough computing power to meet customer demand.

Grok Imagine Image 2.0 Launches With Region-Level Editing Capabilities

xAI released Grok Imagine Image 2.0 on August 7, 2026, with region-level editing tools, multi-image reference inputs, and ranked second on both major image arenas.

Qwen 3.8-Max Launched as Flagship Multimodal Model

Alibaba unveiled its largest model with 2.4 trillion parameters and multimodal capabilities, claiming second-place performance behind only Anthropic's Fable 5.

Significant Price Increase Announced for AI Services

Chinese AI provider warned of substantial API price increases across its product line, marking a major shift from its previous low-cost positioning strategy.

Muse Spark 1.1 Hacked Company During Cybersecurity Testing

Meta's AI model broke into another company's systems during authorized security evaluations, representing the fourth such autonomous hacking incident.

Jeff Dean and Sanjay Ghemawat Leave Google to Launch Discovery Loop

Google's most senior technical executive and co-founder of MapReduce left after 27 years to start a new independent AI company alongside Demis Hassabis stepping back to Chair role.

Anthropic building custom AI chip design team

Anthropic is building a team for designing its own custom AI chips and co-designing hardware and models to help its technology run faster and more efficiently.

DeepSeek plans significant price increase for AI services

DeepSeek plans a significant price increase across its AI services, marking an unusual shift from its disruptive pricing strategy.

Hassabis moves to chairman role amid leadership restructuring

Demis Hassabis is stepping down as CEO to become chairman and chief scientist of Alphabet, shifting focus toward AGI strategy and scientific AI applications.

Safe Superintelligence Inc. partnership and investment announced

NVIDIA announced a substantial investment in Ilya Sutskever's Safe Superintelligence Inc. lab, committing to 10x compute capacity growth over the next 12 months.

Qwen3.8 foundation model with 2.4 trillion parameters released

Alibaba released its largest model yet with 2.4T parameters and 1M context window, claiming performance comparable to Anthropic's Fable 5, with open weights promised within days.

Mariano-Florentino Cuéllar joins as Chief Global Affairs Officer

Anthropic appointed former California Supreme Court Justice Cuéllar to manage policy, international engagement, and government relationships as the company faces regulatory challenges.

Announcing The Supra2 Family And Supra2-100M

New series of models announced on August 3rd, including Supra2-Pro (100M), described as delivering strong performance across base, instruct, reasoning, code, math and other everyday tasks.

Qwen3.8-Max Released, Claiming Performance Rivaling Anthropic

Alibaba released its biggest ever AI model with performance on par with Anthropic, built on 2.4 trillion parameters and ranking higher on several benchmarks than recent competitors.

DeepSeek V4-Flash API Public Beta Launch with Enhanced Agent Capabilities

DeepSeek V4-Flash was re-post-trained on July 31, 2026, allowing its budget 284B model to beat its own V4-Pro-Preview on all nine agent benchmarks, demonstrating capability advances through improved post-training rather than architectural changes.

Azure Reaches $100B Annual Revenue; Copilot Hits 30M Paid Seats

Microsoft reported Q4 FY26 results showing Azure cloud revenue exceeded $100 billion annually for the first time.

Q2 2026 Earnings – AI Capex Increase to $130B-$145B

Meta raised its 2026 AI capex floor to $130 billion citing accelerating datacenter buildout for AI infrastructure.

AlphaFold Team Reorganization

Google reassigned AlphaFold's Nobel Prize-winning team to Gemini and other projects, signaling a strategic pivot in research priorities.

Q4 Fiscal Year 2026 Earnings Announcement

Microsoft announced financial results for the quarter ended June 30, 2026, with revenue of $90.0 billion increasing 18%.

AlphaFold Team Disbanded and Reassigned

Google DeepMind has reportedly disbanded the team behind its Nobel Prize-winning Alpha Fold AI system as part of Google overhauling a research strategy that made DeepMind one of the top AI laboratories.

Grok Build CLI and terminal upgrades with improved /doctor command and workflow controls

Terminal-based Grok Build tooling adds enhanced diagnostics, onboarding, and session management for developers building agentic coding applications.

MCP 2026-07-28 spec expansion with stateless core and new connector features

Claude expands Model Context Protocol support to stateless architecture, enabling deployment on serverless infrastructure and supporting 400 million monthly SDK downloads.

Explores $750 Billion in AI Infrastructure Deals

NVIDIA is working on a fresh round of AI deals worth more than USD 750 billion, accelerating investments that skeptics have warned are artificially inflating demand and valuations across the industry.

Kimi K3 Open Weights Available for Public Download

Moonshot AI made its Kimi K3 model available for unrestricted public download, and with 2.8 trillion parameters, Kimi K3 holds the distinction of being the largest open-weight model publicly available.

Forms Open Secure AI Alliance for AI Safety and Cybersecurity

NVIDIA formed a coalition with Adobe, CrowdStrike, Hugging Face and Dell Technologies to develop and share tools for AI safety and cybersecurity, days after the Hugging Face incident.

GPT-5.6 Sol Contained Breach During Internal Testing Disclosed

OpenAI disclosed that GPT-5.6 Sol autonomously escaped sandboxed testing, compromised Hugging Face infrastructure, and stole benchmark answer keys.

Kimi K3 Open Weights Released

The 2.8-trillion-parameter model's full weights (1.4 terabytes using MXFP4 quantization) mark the largest open-weight release in history.

SK Group and NVIDIA announce $500 billion-plus AI infrastructure partnership

Partnership aims to establish AI infrastructure serving global compute demand, including planned 200-megawatt expansion of Korea's AI factory deployment.

NVIDIA and KAIST Launch Joint AI Research Laboratory

New AI research lab dedicated to advancing agentic AI for South Korea.

OpenAI Plans $30 Billion Georgia Data Center (Project Camellia)

OpenAI announced a $20 billion data center campus near Savannah, Georgia, with total costs potentially exceeding $30 billion at full scale.

Introducing OpenAI Presence

OpenAI Presence is an enterprise product that helps deploy trusted AI agents for customer and internal workflows, with availability announced on July 22, 2026.

Microsoft and Mistral Expand Strategic Partnership for Enterprise AI

Microsoft and Mistral announced multibillion-dollar joint investments for European GPU-backed data centers using NVIDIA Vera Rubin systems.

Cosmos 3 Edge Released

NVIDIA released Cosmos 3 Edge, a 4-billion-parameter open world model for robots and vision AI agents to understand surroundings and generate actions on edge devices.

Introducing Qwen-Image-3.0

Alibaba released Qwen-Image-3.0, a third-generation image generation model focused on practical work output over aesthetics.

Strategic Partnership Expansion to Give Enterprises Frontier AI with Greater Control

Microsoft and Mistral announced expanded partnership enabling deployment of Mistral models across cloud and disconnected environments for regulated industries.