-
industry · OpenAI
Rapidly scaling online storage to serve over 1 billion ChatGPT users (Part II)
OpenAI published Part II of its series on scaling online storage to serve over 1 billion ChatGPT users. It describes how Habitat was scaled with reliable multi-tenancy, global caching, and Azure Cosmos DB to deliver 12× faster reads.
The combination of multi-tenancy, global caching, and a managed database to cut read latency at scale offers a reference architecture for developers designing user-data storage layers or evaluating managed stores like Cosmos DB.
-
release · TechCrunch AI · 2건
Canva releases its first image generation model
Canva released Canva Craft, an image generation model that competes with those from OpenAI, Microsoft, and Google. It uses a two-stage architecture that plans a design before generating the image, supports up to 2K resolution and more than 30 aspect ratios, and is available to all users.
Developers choosing image generation tools can test Canva's claim that plan-then-generate delivers more precision, consistency and control at lower cost against existing OpenAI and Google models.
-
release · Simon Willison
Introducing Kimi K3.1 Pro
Kimi K3.1 Pro is live on the Kimi API, and Moonshot promises full model weights will be released by Nov 4, 2026. It costs $2.00/million input and $10/million output, the same as GPT-6.1 Sol and half of Claude Opus 5.5, and Artificial Analysis scored it 55, behind only Opus 5.5 at 58.
A soon-to-be open-weight model near frontier scores at half Opus 5.5's price lets you reassess model swaps and self-hosting for cost-bound workloads before the weights even ship.
-
research · Hacker News
LLMs are making formal software verification feasible at a much larger scale
A Hacker News post argues that LLMs are making formal software verification feasible at a much larger scale. The title points to LLMs widening how much software verification can cover.
If LLMs make formal verification practical at scale, developers can start weighing proof-based checks alongside code review and testing in their quality assurance design.
-
research · Google Research
Bringing foundational models to the front lines of TB screening
Google Research describes bringing foundational models to the front lines of tuberculosis (TB) screening. The title points to applying these models in frontline health settings.
Developers building medical imaging AI can use it as a reference for how general foundation models might be adapted for frontline screening deployments.
-
industry · OpenAI
Securing our frontier model training environment
OpenAI explains how it strengthened containment for frontier model training and evaluation. The measures include offline environments and high-compute red-teaming.
Teams that train, evaluate, or run agents on their own models can use offline environments and large-scale red-teaming as a benchmark for isolation design and for auditing network access in their evaluation pipelines.