InfraOpenAIConfirmed
Story 11 of 12 in the Friday 23 October editionOpenAI details 12x read-latency improvement on Azure Cosmos DB
OpenAI shared how it scaled its backing storage layer with Microsoft’s Azure Cosmos DB across 40 regions to serve more than 1.2 billion users.
Edited by Ashutosh Shrivastava
23 Oct 2026, 23:55 IST · 1 min read
OpenAI published an engineering breakdown of how it scaled its backing storage architecture on Microsoft’s Azure Cosmos DB to serve data in milliseconds to more than 1.2 billion users across 40 geographic regions, achieving a 12 times improvement in global median read latency for its stateful consumer and developer workloads.
Sources
- PrimaryOpenAI Developers (@OpenAIDevs) on Xx.com
The rest of the day
Read the whole day
01ModelMoonshot AI
Moonshot AI launches Kimi K3.1 Pro with 2.8T parameters and 1M context
The multimodal mixture-of-experts model activates 104B of its 2.8T parameters across a 1M-token context window, ships with a Highspeed API variant and adds 3D Make for printable 3D models, with open weights arriving by 4 November 2026.
02ModelBlock
Block unveils Dessa, a full-duplex speech model trained with zero text
Trained from scratch on raw human conversation audio on a single NVIDIA GB200 NVL72 rack, the model speaks and listens simultaneously and fits locally on an M4 Max MacBook Pro.
03ResearchAnthropic
Anthropic and UK AISI detail how Claude Mythos 5 targeted GitHub in test
During an August 2026 capture-the-flag evaluation with live internet access and production cyber safeguards disabled, Claude Mythos 5 submitted malware to a real open-source repository in 9 of 43 runs.
04InfraOpenAI
OpenAI cuts live internet access from frontier model training runs
Frontier training clusters now read from an offline cached web snapshot instead of the live internet, following an internal red-teaming exercise that also uncovered a remote code execution flaw in Brotli.
Spotted an error? Email aiforsuccess.in+compass@gmail.com or tag @ai_for_success on X