LLM News & AI Model Updates
The latest large language model news: new model releases, benchmark results, pricing changes, and open-source LLM updates, tracked continuously and dated.
What We Cover Here
This hub tracks the large language model landscape as it happens: frontier and open-weight model releases, benchmark score changes, API pricing moves, and tooling launches around coding agents. Every item is dated and sourced, and the current state of the field is always reflected in the AI leaderboard. For benchmark definitions and score histories, see the benchmarks guide.
Latest LLM Updates
Omen Alpha: OpenCode's New Mystery Model Looks Like a Zhipu GLM
Omen Alpha is OpenCode's newest mystery model. Tokenizer clues, provider traces and Reddit reports point strongly toward Zhipu's GLM family.
DeepSeek Harness Explained: Architecture, Plugins, Trajectories & Setup
DeepSeek open-sourced DeepSeek Harness (dsh), a modular plugin-first agent runtime. Here is how it works, its Cordis architecture, and how to run it.
GLM-5.3 Is Live in ZCode: Free Tier, Quota Resets & Agent Benchmarks
Z.ai rolled out GLM-5.3 to all ZCode users with free tier access, reset quotas, and top scores on CyberGym (84.5%), GDPval-AA, and Terminal-Bench 2.1 (88.2).
NVIDIA Nemotron 3.5 Lightning Released: 4x Faster Open AI Agent Model
NVIDIA Nemotron 3.5 Lightning is out now! Discover its 30B/3B-active MoE architecture, 1M context, benchmarks, 4x faster execution speed, and local Ollama setup.
Zed Pro Free Trial: $20 AI Credits, Models & Pricing
The Zed Pro 14-day trial includes $20 in AI credits and unlimited edit predictions. See supported models, costs, limits, and the student plan.
DeepSeek V4 Flash 0731: Pricing, Features, Context Window & Benchmarks
DeepSeek V4 Flash 0731: current pricing, 1M context, features, agent benchmarks, arXiv research paper, and what changed from the preview build.
Muse Spark 1.2: Features, Pricing, Context Window & Use Cases
Muse Spark 1.2 is Meta's coding-focused AI model. See its pricing, 1M context, benchmarks, features, and best use cases.
Why a Dated News Hub Instead of Daily Pages
We deliberately publish news updates into one continuously maintained hub rather than a new page per day. It keeps every update in one place, keeps the record searchable, and avoids polluting the web with thin duplicate pages. When a news item develops into a full analysis - benchmark deep-dives, pricing breakdowns, head-to-head tests - it becomes its own dated article and is linked from here.