[Dev Weekly #117] Open Models Hit Frontier Parity at Half the Cost | AI Agents Go Enterprise | Qualcomm’s $4B AI Bet

by Gabriel Quaresma on July 3, 2026 3 min
Table of Contents

HELLO EVERYONE!!! It’s July 3rd, 2026, and you are reading the 117th edition of Codeminer42’s tech news report. Let’s check out what the tech world showed us this week!

Qualcomm to buy startup Modular for $4 billion in AI software push

Qualcomm is acquiring AI software startup Modular for $4 billion to strengthen its artificial intelligence capabilities. The deal marks a major strategic push into AI software development by the chip giant. Reporting by Reuters highlights this as a significant move in the ongoing AI infrastructure race.

GLM-5.2 vs Claude Opus: Same Code, Less Than Half the Cost

Entelligence AI’s benchmark shows GLM-5.2 matches Claude Opus in coding-agent capability, solving 25 of 45 identical tasks with nearly identical failure modes. The key advantage is cost: GLM-5.2 runs at roughly 46% of Opus’s price when using prompt caching, despite requiring more turns to reach the same results. This indicates open-weight models have reached frontier parity for real-world agentic coding work at less than half the spend.

Agent identity: a new access model for autonomous, team-wide AI — by Noah Zweben

Claude Tag introduces an agent identity model that replaces per-user permissions with workspace and channel-level access controls, letting AI agents operate autonomously in shared team spaces without impersonating individual users. If you’re building multi-user AI systems, this shows how to balance broad tool access for usefulness with scoped permissions for security, plus concrete patterns for configuring roles and audit trails at enterprise scale.

Daybreak: Tools for securing every organization in the world

OpenAI’s Daybreak initiative launches tools like Codex Security and an updated GPT-5.5-Cyber to shift focus from merely discovering vulnerabilities to reduce the burden on security teams and accelerate real-world risk reduction globally.

GLM-5.2 – How to Run Locally

Z.ai’s new 744B parameter GLM-5.2 model is now runnable locally via Unsloth’s Dynamic GGUFs, with quantized versions fitting on consumer hardware (2-bit at 239GB). Learn how to deploy this Claude 4.8 Opus-competitive model on MacOS, Windows, or Linux using Unsloth Studio or llama.cpp, including optimization strategies for reasoning tasks and long-context inference.

Sakana Fugu: Multi-Agent System as a Model

Sakana released Fugu, a multi-agent orchestration system that dynamically coordinates specialized models through a single API to tackle complex reasoning, coding, and research tasks. It achieves frontier-level performance while letting you control which model providers participate, sidesteps single-vendor lock-in, and comes grounded in ICLR 2026 research on learned model coordination.

Nginx will finally make sense after this video

This video explains how Nginx solves the C10K problem by replacing the inefficient thread-per-connection model with an asynchronous, event-driven architecture.

—

And that’s all for this week! Wish you all a great weekend and happy coding!

We want to work with you. Check out our Services page!