Meta Muse Launch, OpenAI Solves Millennium Prize Problem, and More

AI

Meta Launches Muse, a Personal AI Agent

Meta launched Muse on September 8, a personal AI agent that connects to users’ email, calendar, and payment systems to handle scheduling, shopping, and form-filling on their behalf. Unlike conventional chatbots, Muse performs actual tasks rather than just answering questions.

Muse runs inside Muse Secure VM, a dedicated secure virtual machine with its own browser that interacts with apps on the user’s behalf. A free tier is available alongside Power ($20/month) and Maximum ($100/month) paid plans. Access is available via muse.ai, iOS and Android apps, and WhatsApp.

Later in 2026, Meta plans to introduce Muse Confidential VM, which encrypts the entire virtual machine with a key held only by the user. The launch is limited to US users aged 18 and above, with no confirmed timeline for international expansion.

OpenAI Claims Navier-Stokes Millennium Prize Solution

OpenAI announced that it deployed roughly 10,000 AI agents over 88 hours to solve the Navier-Stokes existence and smoothness problem, one of seven Millennium Prize Problems. Each problem carries a $1 million prize from the Clay Mathematics Institute.

The model found that the Navier-Stokes equations can develop a singularity where a vortex tightens and spins ever faster while the fluid’s energy remains bounded. This addresses a 200-year-old question in fluid dynamics. OpenAI said it will not claim the prize money.

The announcement arrived amid a credit dispute. Anthropic employee Levent Alpoge and NYU math professor Tristan Buckmaster had prior work on the problem. OpenAI acknowledged it began its effort on September 1 after hearing a rumor connected to their research.

DeepSeek V4.1 Flash Beta Goes Live

DeepSeek opened a limited beta of its V4.1 Flash multimodal model on September 8, with an official release expected around September 10. This is the largest architecture-level update to the V4 series since its April 2026 debut.

Where the earlier V4-Flash-Vision-Exp bolted a vision extension onto the text model, V4.1 Flash integrates text, image, and speech inputs natively into a unified architecture. Internal and external testing shows the model surpasses V4 Pro across performance, cost, and speed metrics.

The beta is capped at 20 concurrent requests per account. A new pricing structure for the Flash series is scheduled to take effect alongside the official launch.

Big Tech

Qualcomm and AWS Forge Custom AI Inference Chip Partnership

Qualcomm and Amazon announced a multi-year collaboration on September 8 to co-develop custom AI inference chips and 1.6 Tbps optical connectivity for AWS data centers. The partnership spans multiple product generations.

Amazon received warrants to purchase up to 25 million Qualcomm shares at $161.26 each, with full vesting tied to chip, networking, and manufacturing purchases reaching up to $60 billion through September 2036. Qualcomm stock jumped 10% on the news.

This is Qualcomm’s first deal with a major Western hyperscaler, establishing a new competitive axis in the NVIDIA-dominated AI chip market. Qualcomm CEO Cristiano Amon said AI demand acceleration “will require advances in both computing and connectivity to deliver greater performance with more efficiency.”

Chrome 153 Patches Seventh Zero-Day of 2026

Google fixed CVE-2026-87491 in Chrome 153.0.8010.36, an out-of-bounds write vulnerability in the V8 JavaScript engine. An attacker could exploit it via a specially crafted HTML page to execute arbitrary code inside Chrome’s sandbox.

Jihyeon Jeong, a research intern at Seoul National University’s CompSec Lab, reported the flaw and received a $2,500 bounty. Patched versions for Windows, Mac, and Linux shipped two days after the report.

This marks the seventh actively exploited Chrome zero-day in 2026, following CVE-2026-2441, CVE-2026-3909, CVE-2026-3910, CVE-2026-5281, CVE-2026-11645, and CVE-2026-85046. V8 remains a persistent attack surface.

Semifive Ships First Samsung 4nm AI Inference Chip at Scale

South Korean fabless design platform Semifive began mass production of HyperAccel’s LLM inference accelerator “Verda” on Samsung Foundry’s 4nm process. This is the first large-scale production run on Samsung’s most advanced node.

The chip is a 500mm2+ big-die design using an LPDDR-based low-power architecture optimized for LLM inference energy efficiency. Semifive provided a turnkey solution covering frontend design, verification, packaging, testing, and production supply.

Production was originally targeted for Q1 2026 but slipped by two quarters. The milestone is significant for Samsung Foundry as a reference case for its leading-edge process.

Startups

Cognition AI Raises $2B at $48B Valuation

Cognition, the company behind AI coding agent Devin, raised $2 billion in a Series E round at a $48 billion valuation. That nearly doubles its $26 billion price tag from just four months earlier. Andreessen Horowitz and Accel co-led, with Founders Fund, General Catalyst, and NVIDIA among participants.

Annualized revenue has grown from $492 million in May to nearly $900 million. Devin works from high-level requirements, autonomously planning tasks, writing and testing code, debugging, and deploying within a sandboxed environment. NVIDIA, GE Aerospace, Citi, and Mercedes-Benz are among its customers.

Recent product launches include Devin Auto-Triage for incident investigation, Devin Security Swarm for vulnerability detection, and Devin Automations for event-driven workflows triggered from Slack, GitHub, and Linear. Outside of foundation model companies, Cognition ranks among the most richly valued AI startups.

XPeng Commissions IRON Humanoid Robot Production Line

XPeng officially commissioned its IRON humanoid robot production line in Guangzhou on September 8. The facility features over 80% core process automation built to automotive-grade quality standards. IRON became the first humanoid robot to autonomously walk off a production line.

IRON packs 76 degrees of freedom (21 per hand), three Turing AI chips delivering up to 2,250 TOPS, and runs XPeng’s physical AI foundation model on-device. Mass production is set to begin by year-end, with an official launch in China and overseas in 2027.

With Tesla’s Optimus facing production delays, XPeng’s ability to leverage automotive manufacturing infrastructure to reach production readiness first is noteworthy. It represents a tangible milestone in the EV-to-robotics transition.

Open Source

VoiceStudio Trends on GitHub as a Local ElevenLabs Alternative

debpalash/VoiceStudio has climbed to the top of GitHub’s daily trending charts with over 18,000 stars. The desktop suite integrates 16 TTS engines and 11 ASR engines, positioning itself as a fully local, open-source alternative to ElevenLabs.

It supports voice cloning, video dubbing, dictation, transcription, and audiobook creation across 646 languages. A speaker’s voice can be cloned from as little as 3 to 15 seconds of reference audio, and a Voice Design engine can synthesize custom voices from descriptive prompts specifying age, accent, pitch, and emotion.

Everything runs locally on consumer hardware with no external network dependencies. VoiceStudio provides a desktop interface, local API, and MCP service. For users who prioritize privacy and cost control, the value proposition over commercial voice services is clear.

Google Removes uBlock Origin from Chrome; Firefox 155 Stands Alone

Google removed uBlock Origin from the Chrome Web Store on August 31 as part of the final Manifest V2 extension purge. This concludes a phase-out that started restricting the extension in July 2025.

Firefox 155 is now the only major browser that fully supports Manifest V2 and uBlock Origin. Mozilla continues to support Manifest V2, and developer Raymond Hill maintains the complete, unrestricted version for Firefox, including the real-time filtering that Chrome and Edge spent six years walking back.

Chrome users are left with the limited uBlock Origin Lite. For anyone who considers robust ad blocking a core browser feature, the case for switching to Firefox has never been stronger.

Gaming

Don’t Nod Faces Cash Crisis, May Run Out by November

Life is Strange developer Don’t Nod disclosed a severe financial crisis. Cash stood at EUR 8 million at the end of July 2026, with H1 2026 revenue of EUR 6.1 million marking a 56% year-over-year decline. Operating EBITDA loss was EUR 4.3 million.

A string of commercially disappointing releases, including Lost Records: Bloom & Rage (February 2025) and Aphelion (April 2026), is at the root. Major investor Tencent declined to provide further funding. Up to 90 positions in France are at risk under contemplated restructuring.

Auditors warned that the studio’s continuation past January 31, 2027, is uncertain without external financing. The situation highlights a broader pattern in the mid-tier games industry: critical acclaim alone does not guarantee commercial viability.

Virtua Fighter 5 R.E.V.O. World Stage Launches

SEGA’s Virtua Fighter 5 R.E.V.O. World Stage released on September 10 for PS5, Steam, and Xbox Game Pass. It retains the series’ technical, movement-focused fighting system while adding modern competitive features.

Crossplay and rollback netcode are included, addressing baseline expectations for current fighting games. A new solo World Stage mode offers progressively challenging opponents. Upgraded training tools and replay functionality round out the additions.

The release marks a meaningful return for the Virtua Fighter series, with the competitive fighting game community watching closely. Xbox Game Pass subscribers get immediate access at no additional cost.

References

Introducing Muse: The World’s First Personal AI Agent Built for Everyone — Meta
On the Navier-Stokes Millennium Prize Problem — OpenAI
AI Has Solved One of Math’s $1 Million Millennium Prize Problems — Quanta Magazine
OpenAI’s historic math solution overshadowed by credit controversy — Axios
DeepSeek begins limited-time beta of V4.1 Flash multimodal model — TechNode
DeepSeek V4.1 Flash Lightning Beta — KOCPC
Qualcomm lands Amazon deal to build custom AI data center chips — Quartz
Qualcomm Wins First Western Hyperscaler: AWS Deal Pays Up to $60B — Tech Times
Chrome 153 Patches Seventh Zero-Day of 2026 — SecurityWeek
Google patches seventh Chrome zero-day exploited in attacks this year — BleepingComputer
세미파이브, 4나노 반도체 첫 양산 — THE ELEC
세미파이브, 삼성 4나노 기반 AI 가속기 대규모 양산 — Hankyung
Do it all with Devin: Announcing our Series E — Cognition
AI coding startup Cognition raises $2B at $48B valuation — SiliconANGLE
XPENG IRON Humanoid Robot Now Walks Off the Production Line — XPeng
Xpeng opens Iron humanoid robot production line — CnEVPost
XPeng starts IRON humanoid robot production as Tesla Optimus stalls — Electrek
debpalash/VoiceStudio — GitHub
Firefox is now the last major browser that still supports uBlock Origin — PCWorld
Firefox 155 Adds Two New Changes for Faster Web Browsing — It’s FOSS
DON’T NOD Update on business performance for H1 2026 — Don’t Nod Investor Relations
DON’T NOD Is Burning Through Its Last Cash — Game World Observer
Don’t Nod Entertainment Shares Update Regarding Financial Crisis — mxdwn Games
Virtua Fighter 5 R.E.V.O. World Stage — Steam
Virtua Fighter 5 R.E.V.O. World Stage review — TechRadar