SShortSingh.
Back to feed

Z.ai Releases GLM-5.3 Weights With 756GB Package and Revenue-Based License Tier

0
·1 views

Z.ai released the open weights for its GLM-5.3 model on Friday, distributing 756GB of files across 141 Safetensors shards with support for vLLM and SGLang. The model shares the same base architecture as GLM-5.2, with performance gains driven entirely by post-training techniques including reinforcement learning and domain-specific data for software engineering and vulnerability research. Benchmark scores improved significantly on Terminal-Bench, DeepSWE, and ExploitBench, though these were vendor-run evaluations and independent results may vary. The accompanying license allows free use for individuals, startups, and internal enterprise deployments, but requires a Z.ai security review for cloud platforms with group revenues exceeding ten billion dollars before they can offer GLM-5.3 inference or fine-tuning APIs. For developers not running their own infrastructure, Cloudflare has already added GLM-5.3 to its Workers AI platform at launch, priced at $1.40 per million input tokens and $4.40 per million output tokens.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

GKE Launches VPA Decision Logs to Improve Kubernetes Autoscaling Visibility

Google Kubernetes Engine (GKE) has launched a Public Preview of Vertical Pod Autoscaler (VPA) Logs, available on clusters running version 1.36.0-gke.1601000 or newer. Previously, VPA decisions were difficult to audit because standard Kubernetes events expired after roughly an hour, leaving engineers unable to diagnose unexpected pod evictions or failed in-place resizes. The new feature streams structured VPA decision events directly into Cloud Logging under the destination container.googleapis.com/vpa-controller, creating a permanent audit trail. Logs are categorized across four operations — updating recommendations, evicting pods, applying recommendations on eviction, and applying recommendations in place — each tagged with a success, skipped, or failed status. Combined with existing Horizontal Pod Autoscaler logging, GKE operators now have unified visibility across both horizontal and vertical autoscaling dimensions.

0
ProgrammingDEV Community ·

How to Structure an Elixir Project Using Mix: A Practical Guide

As Elixir projects grow beyond a few modules, managing code in IEx sessions becomes impractical due to redefinition warnings and name collisions. Mix, Elixir's built-in build tool, solves this by generating a standard project skeleton that handles compilation, testing, and dependency management. A typical Mix project organizes application code under lib/, with file paths and directory names mirroring module names — for example, LearningElixir.TodoList maps to lib/learning_elixir/todo_list.ex. Non-code assets go in priv/, while config/ separates environment-specific settings for dev, test, and prod. Although the Elixir compiler does not enforce these conventions, virtually all Elixir projects follow them, making the codebase easier to navigate and maintain as it scales.

0
ProgrammingHacker News ·

Opinion: The Modern Internet Has Become Exploitative and Toxic

A blog post by Stephen Diehl argues that the contemporary internet has devolved into a predatory environment for users. The piece, shared on Hacker News, attracted 26 upvotes and sparked discussion among readers. Diehl's essay examines how online platforms and digital ecosystems have shifted toward exploitative practices. The post reflects growing public concern about the commercialization and manipulation embedded in today's internet experience.

0
ProgrammingDEV Community ·

Why Organisational Constraints Often Matter More Than Technology Choices

Technology challenges in organisations rarely stem from technical limitations alone, according to observations from advisory firm Cralgo. Slow delivery, shifting priorities, and poor decision-making are frequently rooted in unclear ownership, misaligned teams, and weak governance rather than inadequate tools or platforms. Investments in cloud, AI, and automation only deliver value when the organisation surrounding them can effectively direct and use that capability. As companies scale, informal context-sharing breaks down, causing the original intent behind decisions to erode before reaching execution. The critical gap, the article argues, lies in the connective layer between strategy and delivery, where sound judgement must be carried consistently into action.

Z.ai Releases GLM-5.3 Weights With 756GB Package and Revenue-Based License Tier · ShortSingh