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.
The 689M parameter model supports 91 languages and processes documents at 4.5 pages/s on a single H100 GPU.
Apache 2.0 licensed tool processes 491 questions per second on an RTX 4090 by reading content once for multiple decision queries.
The new models use full-image inference without zooming tools, improving overall accuracy from 55.52 to 60.51 across six domains.
The open-source tool provides a GUI for organizing AI agent workspaces, tracking agent status, and keeping terminals running in the background across local and remote machines.
JetBrains released ThinkRail v0.2.0, an IDE with the Pi coding agent embedded.
The family includes reference-based and reference-free variants released on Hugging Face under OpenMDW-1.1 and Apache 2.0 licenses.
The largest 1B-parameter model achieves 44.71% average zero-shot QA accuracy, surpassing the 39.79% of the corresponding GPT baseline.
The new display serif shares metrics with Geist Sans and includes a variable axis for swapping dot shapes between round, square, triangle, and heart.
The ~70 MB model uses 200+ trained heads on a shared encoder to output transcripts, linked names, and metadata every 100 ms.
Swift GLM-5.3-Flash achieves -61.4% thinking tokens on MMLU-PRO with -0.46% average accuracy loss across nine benchmarks.
The tool indexes local Claude Code session histories to identify broken workflows and silent regressions, such as agents failing to load skills due to edited descriptions.
The patch release updates the Granite Speech ASR model for 10 languages and improves the Decision adapter without recomputing the base model.
SafeAgents is an open-source local proxy that uses up to three checks, including 110 deterministic detectors, to intercept and block malicious or out-of-scope tool calls before execution.
The 27B parameter open-weights model is free for non-commercial use and outperforms real human writing in detector evasion tests.
tstack, released by Theo Browne, is a set of 9 skills that enable coding agents to handle the process up to PR merging.
The open-source plugin targets gfx1151 hardware to optimize MoE GEMMs and attention layers, reducing 256k prompt prefill time from over two hours to four minutes.
The release supports 53 CLI adapters and uses a deterministic Python scheduler to ensure byte-identical replay of agent plans without LLMs in the coordination loop.
CopilotKit has open-sourced OpenDuet, a self-hostable proactive work agent.
The packages wrap Apple's new SpeechAnalyzer API with NVIDIA and Alibaba models, achieving 1.55% WER on LibriSpeech test-clean, and enable typed decision outputs via six backends including MLX and llama.cpp.
Liquid AI has released six 2B-scale LFM3 models supporting execution on edge devices and multimodal processing.
The 1B OCR model runs locally on mobile devices while the 2B verifier model checks output accuracy against original images.
Open-weight model supports 40 languages and speaker diarization, with a 30B parameter architecture using 4B active parameters per token.
The model features a 50M parameter denoiser and generates images in 11 seconds at 20 steps on an Intel i5 7200U.
The model features a 35B-parameter mixture-of-experts backbone with approximately 3B active parameters per token and a 128K context window.