Discover the Best AI Tools

We test the top AI tools for writing, video, images & music — so you don't have to.

Explore AI Tools

OpenAI's Decisions API Is Now Live for Everyone — and It Was Built to Catch a Startup

Image
  Most AI answers arrive as essays. OpenAI's new Decisions API doesn't write a word — it answers yes or no, picks from your categories, or scores on your rubric, in 100 to 180 milliseconds. It went generally available this week, and the backstory is the best part: OpenAI's own product lead, Nikunj Handa, reportedly admitted the product wasn't on the roadmap four weeks ago. A small startup forced the sprint. Yes or no, in 100 milliseconds The API runs on GPT-6 Luna. Developers send text or images plus typed questions and get machine-readable answers in three shapes: a probability that a condition is true, a pick from a fixed list with a probability for each, or a score on rubric levels you define. Input costs $0.10 per million tokens, and output tokens are free. It launched as a public beta on October 6 (after a limited DevDay preview in late September), and the general release adds voice and visual input plus geographically targeted inference. Early customer Speak, the...

OVHcloud Is Training Its Own AI Model on Europe's Fastest Supercomputer

OVHcloud AI model cover: giant AI robot head over European supercomputer data hall

 

OVHcloud logo


Mistral logo

A cloud provider just became a model maker. OVHcloud — France's largest independent cloud company — expects to launch its first proprietary AI model in early 2027, CEO Octave Klaba told Reuters on Tuesday after the company reported its annual earnings.


It's a striking move: OVHcloud makes money hosting other companies' workloads. Now it wants to compete with the companies that build the brains.


What the model will be


Klaba gave three concrete details. The model will use a mixture-of-experts design — instead of one giant model handling every request, it routes tasks to smaller specialized sub-models, a trick that cuts inference cost and latency. Training is already underway on Jupiter, which Klaba calls Europe's fastest supercomputer — work that started back in June. And before the model arrives, OVHcloud will pilot OVHai Workspace, a productivity platform tuned on the company's own internal data for everyday office work like email.


Why OVHcloud thinks it can win


The argument is capacity. OVHcloud has 250 megawatts of connected cloud capacity, but only 140 MW is currently in use — leaving 110 MW open for new customers. That means it can onboard AI workloads without spending on new data centers, while competitors must invest heavily just to keep up with regional demand.


Klaba is also positioning OVHcloud as the potential European challenger to Mistral — France's AI flagship — and he isn't shy about his real target. He has spent months criticizing Europe's state-backed computing programs, calling national strategies for the EU's planned AI gigafactories "doomed to failure" and arguing that a proposed 200-million-euro public commitment to buy AI services is far too small to attract the private investment the gigafactories need.


His message: Europe doesn't need more government blueprints — it needs companies that already have the megawatts and the customers.


The sovereignty angle


This isn't happening in a vacuum. Europe has spent years talking about AI sovereignty — models and infrastructure that keep European data under European control, free from US and Chinese cloud giants. Scaleway is already building its own AMD-powered supercomputer in France, and now a second European cloud company is entering the model business.


If OVHcloud delivers, Europe's AI race shifts from policy papers to price-per-token competition — and the challenger won't be a lab with a famous name, but a hosting company with empty server halls to fill.


My take


Four things have to go right: the Workspace pilot has to land, training has to finish on schedule, benchmarks (still unannounced) have to be competitive, and the company needs a pricing story that undercuts Mistral and the US frontier labs. Watch for a model name and an API preview — silence through mid-2026 would say the timeline is soft. But the trend is unmistakable: the cloud providers are done renting their GPUs to AI labs. They want to sell the intelligence themselves.

Comments

Popular posts from this blog

AI Tools Every Accountant Should Use in 2026

Top 5 Best AI Video Generation Tools in 2026

7 Best AI Website Builders in 2026