Today's AI news: DeepSeek v4.1-Flash, OpenAI's Navier-Stokes milestone, GPT-6 Astra partnerships, AlphaGenome Atlas, and ChatGPT hitting 1 billion users.
Here is your daily digest of the most significant developments in AI for 12 September 2026.
DeepSeek v4.1-Flash Arrives with a Novel Encoder–Decoder Architecture
DeepSeek has released v4.1-Flash, a 763-billion-parameter model featuring a novel causal Encoder–Decoder architecture with vision capabilities. The model uses a parameter configuration described as P8B-D16B and marks what observers are calling the "Return of the Whale" — a significant scaling push from the Chinese AI lab. Commentary from the AI research community, including from Sebastian (cited in the AINews roundup), suggests the release is substantial enough that it arguably should have been versioned as DeepSeek v5 rather than a point release. The addition of vision capabilities and the departure from a standard decoder-only design make this one of the more architecturally distinctive releases in recent memory.
OpenAI Claims Navier-Stokes Singularity Discovery — A Potential Millennium Prize Contender
In what is being described as "the most jam-packed, feel-the-AGI day in the history of AI," OpenAI has reported finding a Navier-Stokes singularity solution — a problem that has stumped mathematicians for centuries and is one of the seven Millennium Prize Problems. According to the AINews report, the result was achieved in 88 hours using a system called Astra-next, approximately 10,000 agents, and around 130 billion tokens of compute at a cost exceeding $40 million USD. If verified, this would be only the second Millennium Prize ever awarded. The announcement overshadowed several other major news items on the same day, including Cognition's $48 billion Series E funding round and Mistral's $24 billion Series D.
GPT-6 Astra Powers Real-World Deployments at Perplexity and Cognition
OpenAI's GPT-6 Astra model is now being deployed in production by two high-profile partners. Perplexity is using Astra to write communications, modify software, and monitor production systems — with the company reporting it checks in on the model far less frequently than it did with earlier models, indicating a meaningful leap in autonomous reliability. Meanwhile, Cognition is using GPT-6 Astra to improve its Devin software engineering agent's ability to test its own work, with the goal of reducing the volume of code that engineers need to review before shipping. Both deployments point to a broader shift toward AI systems trusted with end-to-end operational tasks rather than narrow, supervised subtasks.
AlphaGenome Atlas Maps 9 Billion DNA Variants
Google DeepMind has published the AlphaGenome Atlas, described as a predictive map of every possible single-letter DNA change in the human genome. The atlas covers the molecular effects of 9 billion single-nucleotide variants — a dataset of extraordinary scale that could significantly accelerate genomic research and help scientists understand the impact of genetic mutations on human health. This follows a series of major biology-focused AI releases from DeepMind and continues the lab's focus on applying AI to life sciences at a foundational level.
ChatGPT Reaches 1 Billion Users as OpenAI Scales Storage Infrastructure
OpenAI has published a detailed account of how it scaled its internal storage platform — called Habitat — to serve over one billion ChatGPT users, handling 22 million requests per second. The platform evolved from a Python library into a globally distributed storage system. The milestone reflects the extraordinary growth in ChatGPT's user base and the engineering challenge of supporting that scale reliably. It is also a notable data point for the broader industry: ChatGPT is now the largest deployed consumer AI product by user count.
OpenAI's Data Agent Brings Natural-Language Analytics to ChatGPT Work
OpenAI has launched a Data agent within ChatGPT Work, allowing users to connect company data sources, surface insights, and build interactive dashboards using natural language. The feature is positioned as making data analysis accessible to non-technical users, removing the barrier of needing to write code or use specialised business intelligence tools. This follows a broader trend of AI assistants expanding into structured business workflows rather than remaining purely conversational.
Frequently Asked Questions
What is DeepSeek v4.1-Flash and why is it significant?
DeepSeek v4.1-Flash is a 763-billion-parameter AI model featuring a novel causal Encoder–Decoder architecture with vision capabilities. It is considered significant enough by researchers that some believe it should have been versioned as DeepSeek v5. Its architectural departure from standard decoder-only designs makes it one of the more distinctive model releases in recent memory.
Did OpenAI solve the Navier-Stokes Millennium Prize Problem?
OpenAI has reported finding a Navier-Stokes singularity solution using its Astra-next system. The result was achieved in 88 hours using around 10,000 agents and 130 billion tokens of compute at a cost exceeding $40 million USD. If independently verified, it would be a contender for only the second Millennium Prize ever awarded.
How is GPT-6 Astra being used in industry?
GPT-6 Astra is being used by Perplexity to write communications, change software, and monitor production systems with minimal human check-ins. Cognition is also using it to improve its Devin software engineering agent's ability to test its own code, reducing the amount of code human engineers need to review.
What is the AlphaGenome Atlas?
The AlphaGenome Atlas is a tool released by Google DeepMind that maps the molecular effects of 9 billion single-letter DNA variants across the entire human genome. It is designed to help researchers understand how genetic mutations affect human health at a foundational level.