EVA Tech BriefIT/DevOps

2026-06-26 - 7:30 PM PKT

Frontier LLMs

OpenAI internal Codex usage surges across functions

Latent Space highlighted OpenAI's report that median internal Codex output tokens have grown sharply since November 2025, including 56x in research and 32x in customer support. The numbers point to agentic coding and analysis moving from developer-only workflows into broader operational teams.

Hugging Face makes vLLM serving available through HF Jobs

Hugging Face published a one-command workflow for running a vLLM server on HF Jobs. The post is aimed at teams that want to stand up OpenAI-compatible inference endpoints for open models without building their own deployment plumbing first.

GPUs and Silicon

IBM details 0.7nm-class semiconductor technology

Tom's Hardware reports that IBM has developed a sub-1nm, 0.7nm-class process using nanostack transistors. IBM is claiming major performance and efficiency gains versus its 2nm-class node, which keeps advanced chip scaling relevant for future AI and high-performance compute roadmaps.

Apple may skip high-end M6 chips and fast-track AI-focused M7 Macs

A report covered by Tom's Hardware says Apple may ship a base M6 chip for entry-level Macs while skipping M6 Pro and Max variants. The company is reportedly accelerating an AI-focused M7 generation for 2027, suggesting Mac silicon priorities are shifting toward local AI workloads.

Cloud and Infra

Runpod raises $100M for AI developer cloud

Runpod raised $100 million in Series A funding, bringing its valuation to $1 billion. The company is positioning itself as a developer-first AI cloud for GPU workloads, a category that keeps expanding as teams look beyond hyperscalers for inference and training capacity.

New MCP specification reduces old risks but creates new ones

Akamai researchers reviewed the upcoming Model Context Protocol specification and found that it removes several earlier protocol-level security issues. The same changes introduce new attack surfaces, so teams adopting MCP for cloud and agent integrations will still need explicit threat modeling and controls.

Pakistan Tech

Mobile apps drive Pakistan's digital payments growth

TechJuice reports that Pakistan processed 3.7 billion retail transactions worth Rs168.8 trillion in Q3 FY26, with digital channels accounting for 92 percent of transaction volume. Mobile banking apps and internet banking processed Rs68 trillion, showing continued migration from branch-led banking to app-based payments.

Punjab announces temporary mobile service restrictions in 24 districts

Punjab's Home Department announced localized mobile phone service restrictions around sensitive Youm-e-Ashura processions and Majalis venues. The telecom disruption is temporary and security-driven, but it is operationally relevant for teams that depend on mobile connectivity for field coordination and incident response.

Security

Miasma malware targets npm packages and GitHub Actions

Researchers flagged new supply-chain activity tied to the Mini Shai-Hulud, Miasma, and Hades malware family. The campaign affects npm packages, abuses GitHub Actions workflows, and has signs of spreading into the Go ecosystem, making dependency and CI pipeline review a priority.

Amazon Q Developer flaw exposed cloud credentials through malicious repos

A high-severity Amazon Q Developer vulnerability could let a malicious repository execute commands after a developer trusted the workspace. The patched CVE-2026-12957 issue involved handling of MCP server configs and could expose cloud credentials, underscoring the risk around AI coding tools and repo trust boundaries.

Dev Tools

OpenAI reportedly slows GPT-5.6 rollout for government review

The Verge reports that OpenAI will initially release GPT-5.6 only to a small group of partners after a Trump administration request tied to safety concerns. The staged rollout shows frontier model launches are becoming more controlled and may affect when developers see broad API availability.

OpenKnowledge launches as an open-source AI-first notes tool

OpenKnowledge launched on Hacker News as a local, open-source markdown editor positioned as an alternative to Obsidian and Notion. It includes direct integrations with Claude, Codex, and Cursor, giving developers a docs workspace that connects writing, search, and AI-assisted editing.