Trending

Trained LLMs exclusively on pre-1913 texts

GPT-5.2-Codex

Prompt caching: 10x cheaper LLM tokens, but how?

FunctionGemma 270M Model

Show HN: A local-first memory store for LLM agents (SQLite)

Developers can now submit apps to ChatGPT

Ditch the Chain-of-Thought Hacks: A Modular System for Composing AI Operations

Debug Mode for LLMs in vLLora

Show HN: Solving the ~95% legislative coverage gap using LLM's

Article: NextGen Search - Where AI Meets OpenSearch Through MCP

In this article, authors Srikanth Daggumalli and Arun Lakshmanan discuss next-generation context-aware conversational search using OpenSearch and AI agents powered by Large Language Models (LLMs) and Model Context Protocol (MCP). By Srikanth Daggumalli, Arun Lakshmanan

TornadoVM 2.0 Brings Automatic GPU Acceleration and LLM support to Java

The TornadoVM project recently reached version 2.0, a major milestone for the open-source project that aims to provide a heterogeneous hardware runtime for Java. The project automatically accelerates Java programs on multi-core CPUs, GPUs, and FPGAs. This release is likely to be of particular interest to teams developing LLM solutions on the JVM. By Ben Evans

Podcast: Building a More Appealing CLI for Agentic LLMs Based on Learnings from the Textual Framework

Will McGugan, the maker of Textual and Rich frameworks, speaks about the reasoning of developing the two two libraries and the lesson learned. Also, he shares light on Toad, his current project, which he envisions being a more visually appealing way of interacting with agentic LLMs through command line. By Will McGugan

Podcast: Platform Engineering for AI: Scaling Agents and MCP at LinkedIn

QCon AI New York Chair Wes Reisz talks with LinkedIn’s Karthik Ramgopal and Prince Valluri about enabling AI agents at enterprise scale. They discuss how platform teams orchestrate secure, multi-agentic systems, the role of MCP, the use of foreground and background agents, improving developer experience, and reducing toil. By Karthik Ramgopal, Prince Valluri

OpenAI's New GPT-5.1 Models are Faster and More Conversational

OpenAI recently released upgrades to their GPT-5 model. GPT‑5.1 Instant, the default chat model, has improvements to instruction following. GPT‑5.1 Thinking, the reasoning model, is faster and gives more understandable responses. GPT‑5.1-Codex-Max, the coding model, is trained to use compaction to perform long-running tasks. By Anthony Alford

Replit Introduces New AI Integrations for Multi-Model Development

Replit has introduced Replit AI Integrations, a feature that lets users select third-party models directly inside the IDE and automatically generate the code needed to run inference. By Daniel Dominguez

Using 3 different LLMs to build/code games for a smart ball

An independent Korean researcher is trying to democratize LLM pretraining with a 1.5B model

GPT-5 demonstrates ability to do novel lab work

It was Ilya who "closed" OpenAI

NVIDIA releases Nemotron 3 Nano, a new 30B hybrid reasoning model!

New Google model incoming!!!

llama.cpp: Automation for GPU layers, tensor split, tensor overrides, and context size (with MoE optimizations)

I pitted GPT-5.2 against Opus 4.5 and Gemini 3 in a robot coding tournament

Qwen3-Next-80B-A3B-Thinking-GGUF has just been released on HuggingFace

NVIDIA

Save item

NVIDIA-Nemotron-3-Nano-30B-A3B-BF16

Task: text-generation

FunAudioLLM

Save item

Fun-CosyVoice3-0.5B-2512

OpenAI

Save item

circuit-sparsity

Task: text-generation

Mistral AI_

Save item

Devstral-Small-2-24B-Instruct-2512

Z.ai

Save item

GLM-4.6V-Flash

Task: image-text-to-text

Google

Save item

functiongemma-270m-it

Task: text-generation

NVIDIA

Save item

NVIDIA-Nemotron-3-Nano-30B-A3B-FP8

Task: text-generation

Z.ai

Save item

AutoGLM-Phone-9B-Multilingual

Task: image-text-to-text

Unsloth AI

Save item

Nemotron-3-Nano-30B-A3B-GGUF

Task: text-generation

BlinkDL

Save item

RWKV-LM

RWKV (pronounced RwaKuv) is an RNN with great LLM performance, which can also be directly trained like a GPT transformer (parallelizable). We are at RWKV-7 "Goose". So it's combining the best of RNN and transformer - great performance, linear time, constant space (no kv-cache), fast training, infinite ctx_len, and free sentence embedding.