Daily AI News | Codex reaches 20 million active users and gives every user a banked reset
# Daily AI News | Codex reaches 20 million active users and gives every user a banked reset
## Lead Story
### Codex reaches 20 million active users and gives every user a banked reset
Codex product lead Tibo Sottiaux said Codex reached 20 million active users this week. To mark the milestone, every Codex and ChatGPT Work user will receive a banked usage reset during the day; the one-time credit does not represent a change to subscription prices or permanent quotas.
**AI take:** The reset can extend capacity for demanding work, but it is a one-time reward rather than a permanent limit increase.
## More News
### YMTC parent seeks RMB 33 billion in Shanghai STAR Market IPO
Reuters reported that the Shanghai Stock Exchange accepted the STAR Market application of YMTC parent CCSH Corporation, which plans to raise RMB 33 billion, or about $4.9 billion. The prospectus directs proceeds to production upgrades and next-generation NAND research and implies a post-listing valuation of roughly RMB 275 billion to RMB 330 billion.
**AI take:** If completed, the deal would give YMTC substantial capital for capacity and R&D, though the final size still depends on review and market conditions.
### ChatGPT on Mac can search and send Apple Messages
OpenAI launched an Apple Messages plugin for ChatGPT on Mac, allowing ChatGPT Work and Codex desktop users to search conversations, catch up on context, draft replies, and send iMessage, SMS, and RCS messages. The feature requires macOS permissions for message data, contacts, and automation, with confirmation required by default before sending.
**AI take:** It can reduce context switching, but message history and broad system permissions make narrow, deliberate access essential.
### DeepSeek launches the experimental V4-Flash-Vision-Exp multimodal model
DeepSeek announced that V4-Flash-Vision-Exp is live on its official API platform. The company says the experimental model retains V4-Flash text capabilities across agents, reasoning, and world knowledge while adding visual understanding under a new model identifier.
**AI take:** It brings images into existing V4-Flash workflows, but production stability and long-term compatibility still need real-world validation.
### Doubao Work Tasks adds skills, connectors, and agent teams
Doubao rolled out a skills store, connectors, and work partners or teams in the desktop Work Tasks mode, with more than 200 skills and connectors plus support for custom skills. Authorized connectors can access tools such as Feishu and DingTalk, while specialized agents can collaborate on multi-step work.
**AI take:** The update turns one-off chats into reusable workflows, but account permissions and auditable actions will determine safe adoption.
### Runway launches Ruby for converting short video to up to 16-bit HDR
Runway launched Ruby, a model that converts SDR video to as much as 16-bit HDR and exports ProRes or EXR sequences. It works with uploaded video and Runway-generated output, but each source clip is currently limited to 30 seconds.
**AI take:** Ruby targets professional post-production rather than general generation, and longer projects will still require segmented workflows.
### Starcloud adds $250 million to advance orbital data centers
Starcloud told TechCrunch that it added a $250 million extension to its $170 million March Series A, reportedly valuing the company at $2.3 billion. The capital will expand manufacturing and advance the larger Starcloud-3 orbital data-center spacecraft intended for launch on SpaceX Starship.
**AI take:** The financing shows strong appetite for orbital compute, but commercialization still depends heavily on launch cost, capacity, and thermal management.
### Qianhe Yibang says its four-layer 3D DRAM compute-in-memory chip has powered on
Qianhe Yibang announced that its four-layer 3D DRAM stacked compute-in-memory chip returned from fabrication and successfully powered on, using a logic layer beneath four DRAM layers in a “4+1” design. The company says the 3D integration is intended to raise memory bandwidth and reduce power and latency; the performance figures remain company claims.
**AI take:** Power-on clears an early engineering milestone, but yield, cooling, and volume manufacturing will determine data-center viability.
### Nevada approves commercial robotaxi permits for Tesla, Uber, and Waymo in Las Vegas
The Nevada Transportation Authority unanimously approved permits allowing Tesla, Uber, and Waymo to operate commercial robotaxi services in Clark County, home to Las Vegas. The permits set combined ceilings of up to 8,000 vehicles over the next year—5,000 for Tesla and 1,000 each for Waymo and Uber—while company testimony indicated actual deployment could be far lower.
**AI take:** The decision opens the commercial pathway, but “up to 8,000” is a regulatory ceiling rather than a fleet already on the road.
## Rumors
