AI Briefing#open-source
The latest AI and developer news about #open-source, with the original source and a short summary.
The latest AI and developer news about #open-source, with the original source and a short summary.
Neuphonic has open sourced NeuDecide, a 43 MB audio-to-tool model that achieves 72.4% tool accuracy on SLURP benchmarks without transcription, running efficiently on a single CPU thread.
The initiative introduces the Critical Infrastructure Defense Program with 11 founding partners and a free OSS Scanner service for vulnerability detection.
The 0.8B Jeff model determines whether to think on each turn, reducing Qwen3.8-27B's task time by 32%.
Tencent ARC Releases WorldCrafter: Open-Source Video World Model with Implicit 3D Memory
Lemma has open-sourced a platform that lets teams use their existing AI subscriptions (like Claude Code or Codex) within a shared, permission-controlled workspace with multi-surface access.
Perplexity has released pplx-embed-v2-late, a family of ColBERT-style multi-vector embedding models in 0.6B and 9B sizes, supporting multimodal retrieval for text and visual documents.
It supports isolated multi-user agents within a single process and provides 40 LLM presets and an expert template library.
The dataset covers 106 SLMs with 150M parameters or fewer, evaluated on a Runpod B200 under Apache 2.0 license.
The tool allows developers to define, run, and share multi-agent AI systems using declarative YAML configurations and the `docker agent` CLI plugin.
The new d1-3B model achieves a score of 48.57 on the Decision Index 0.2.1, outperforming larger models like Decider 35B-A3B.
The model achieves a 20.92% average Word Error Rate on Arabic benchmarks, outperforming the previous best published result by 2.25 percentage points.
Nemotron-3-Ultra-CC scored 535.4/600 at IOI 2026, surpassing the top human score, while a specialized IMO system achieved 30/42 points.
rGPU is an open-source tool that enables PyTorch applications to execute operations on remote NVIDIA GPUs while keeping the client application local, supporting users on systems like Macs without CUDA.
The MIT-licensed platform supports multi-user isolation, local-first data storage, and integration with external coding agents like Claude Code and Codex via the Agent Client Protocol.
The Apache-2.0 licensed engine syncs IMAP/JMAP accounts into an event log and exposes actions via HTTP, SSE, webhooks, and MCP without including an internal model.
The Aimee project has released vLLM plugins for Qwen 3.8 and Gemma4 models, enabling native memory access that decouples knowledge from model parameters and eliminates the need for retraining.
Berthd offers an open-source macOS client that runs Claude Code and Codex on remote dev boxes, allowing agents to continue working even when the local laptop is closed.
The new AFP-GIC framework reduces decoder latency by 18.1% and parameter count by 20.5% compared to DC-VIC while supporting five bitrate operating points.
The open-source Terse plugin for Claude Code enforces 22 writing rules to eliminate verbose 'Claude-speak,' with benchmarks showing word count reductions of 46-54% and cost savings of 32-50%.
Using Qwen3-8B on FanOutQA, CacheBack achieves 3.2× faster median task completion and 14.7 percentage points higher accuracy than text-based communication.
Treg acts as an 'OpenRouter for tools,' providing a single base URL and token to access thousands of external endpoints, handling credential injection and billing for AI agents.
The openTPU project presents an open-source AI accelerator developed by AI agents, capable of running models like LFM2.5-230M and Qwen3.5 on a Xilinx Kintex-7 FPGA with bit-exact simulation results.
The 740M-parameter model unifies text, images, audio, and video into a shared embedding space under an Apache 2.0 license.
The Document Foundation states it will not add generative AI to LibreOffice's default configuration, prioritizing local data processing over cloud-based features.