#Open-source ecosystem

AITencent's WeVisDoc-4B, fine-tuned from Qwen3-VL-4B-Instruct, converts page images into structured Markdown with LaTeX formulas and HTML tables. It scores 95.38 Overall on OmniDocBench v1.6 and a mean Overall of 75.54 across three PureDocBench tracks, ranking first among compared end-to-end parsers in all four reported settings. The model is available on Hugging Face and runs through vLLM, which requires version 0.11.1 or later.

AIZed has launched the public beta of Delta, a multiplayer environment for coding with agents and reviewing their work, which replaces pull requests with shared threads. Delta is built on DeltaDB, which records edits and messages between Git commits, and it is free during the beta, with paid plans for individuals and teams to follow.

Why it matters: The post explains how Delta replaces pull requests with shared agent threads and DeltaDB, showing a concrete alternative to the GitHub review workflow.

AITogether AI says teams can migrate from closed to open-source models in weeks to months, not years, using a managed service. The guide recommends defining the use case, shortlisting candidates with relevant benchmarks, and replaying existing traffic against them to test accuracy, cost, and latency against current targets.

AIPeriodic Labs, the team behind high-throughput materials labs in Menlo Park, reports progress on one of condensed matter physics' hardest problems. Its open-source model Neon, trained with mid-training and RL on 1,300 H200s plus months of lab data, surpasses GPT-6 Astra on the company's analysis benchmark. The work targets materials science challenges including superconductors, magnets, and semiconductors.

AIv0 now supports multiple AI models through Vercel's AI Gateway, spanning frontier, cheap, open, and fast options. Users can pick from models including Claude, GPT, Kimi, GLM, Grok, and DeepSeek to build their apps.

AILlamaIndex says Stainless helped it keep LlamaParse SDKs current and pushed it to make the API's names and schemas more consistent. With the Stainless team joining Anthropic, George He and Yong Park explain what worked, what they learned, and why changing SDK generators needs careful handling.

AIGoogle reports that its technologies and products now power everyday interactions in more than 300 languages used by over 7 billion people, about 86% of the global population. The post describes new speech models, including Gemini 3.5 Live Translate and Gemini 3.5 Transcribe, plus the TranslateGemma open translation models trained across 55 languages.

AIHugging Face's Leandro von Werra argues that frontier AI labs should release small variants of their models, share core parts of their alignment recipe, and publish tech reports with more than evaluations. He says these steps would let the wider community test model behavior and verify safety claims, rather than leaving the safety agenda to a few labs. He also calls for independent verification of alarming internal findings, with sensitive details disclosed first to an independent team.

AI🚀 SGLang has Day-0 support for Intern-S2-397B from @intern_lm, a 397B multimodal foundation model built for scientific intelligence and long-horizon agents. New pre-training paradigm: learns directly from raw scientific literature pages, no parsing needed. Scientific reasoning: RL across 20+ domains, from biomolecule design to material generation. Long-horizon agents: black-box agentic RL in large-scale sandboxed environments. Run it now with SGLang!

AINovita AI has open-sourced Chord, a W4A16 MoE CUDA operator with BF16 activations, INT4 weights and group-32 scales, built for Kimi K2.x serving shapes. Measured per layer against public Humming, it reports 1.11–1.20x on H200 EP8 prefill, 1.17–1.33x on H200 TP8 serving, and 1.81–2.15x on B300 EP8 decode against an untuned Humming default. Integration of the grouped operators with vLLM's Humming backend is still a work in progress.

AIPart 2 of the zero-trust agents series moves security checks from agent code to the Gemini Enterprise Agent Platform runtime. Model Armor screens prompts and responses, Semantic Governance Policies judge proposed tool calls against intent and business rules, and Agent Anomaly Detection flags multi-turn drainage that single-turn checks miss. The same Customer Support and Returns Agent from Part 1 is used, with the companion demo open-sourced on GitHub.

Why it matters: The post walks through a concrete refund agent under four attacks, showing how screening, intent judgment, and anomaly detection each catch what the others miss.

AIThe vLLM team trained a DSpark speculative decoding draft model for Kimi K3, a 2.8T-parameter model, using the Speculators library on GB300 NVL72 hardware. They added a MooncakeHiddenStatesConnector to stream hidden states from disaggregated vLLM inference nodes to training nodes across multiple machines. The released speculator raises single-stream interactivity from about 110 to about 435 tokens per second per user on math reasoning, with up to about 3.5x higher output throughput under concurrent load.

Why it matters: The post shows how hidden-state extraction and Mooncake transfers let a 2.8T-parameter model's speculator be trained across multiple nodes, a reusable pattern for similar setups.

AIMiniMax H3 with SGLang-Diffusion and VDN-H3 generates 14.4 seconds of 768p video in 9.0 seconds end-to-end after warmup on 8× B200 GPUs. Eight-step denoising takes 6.9 seconds, exceeding 2× real-time, with no measured quality regression versus dense 50-step H3 across 103 test prompts.

AISenseNova Skills is an open-source agentic office skill suite from SenseTime that combines Deep Research, Data Analytics, PPT Generation, and Infographic Generation. It converts raw data into editable .pptx decks and is built for any skills-compatible agent, including OpenClaw and hermes-agent.

AIIntern Large Models introduces Intern-S2-397B, a multimodal foundation model built for long-horizon scientific research and scientific agents. The post reports leading open-source results on IMO-Proof and AdvancedMathBench, and says the model reaches the level of Gemini 3.1 Pro on those tasks. It is now supported by vLLM and SGLang, with weights on Hugging Face and ModelScope and a chat demo available.

AIMiniMax highlighted open-source community progress on its H3 video generation model, which it built with native stereo audio and multimodal reference control. Recent highlights include FastH3's 4-step distillation running on DGX Spark and Apple Silicon, and NVIDIA's Sol-H3 generating 15 seconds of 768p video with audio in 6.6 seconds on 8×B300 in a warm-inference benchmark. Other releases include VDN's faster-inference attention work with code and weights, and 8-step Acc-LoRAs from Alibaba PAI, with LightX2V offering 4- and 8-step Turbo LoRAs.

AIinclusionAI released Step-3.7-Flash-singprobe, an 8.13M-parameter probe that reuses Step-3.7-Flash hidden states to score query intent, response unsafety, and hallucination risk at every generated token. The probe adds less than 0.5% decode-time overhead and reports 0.9858 R-AUC and 0.9295 T-AUC on streaming safety benchmarks. It is supported through SGLang and vLLM integration branches and loads from Hugging Face by checkpoint ID.

AIIan Johnson visualizes 30 million jina-v5-nano embeddings from 12 billion tokens across multilingual FineWeb, StarCoder, The Pile, and RedPajama as a flyable video game. He frames the project as making data exploration engaging rather than a chore.

AISebastian Raschka's third "Reasoning from Scratch" video covers building a math verifier for evaluating language models and for later reinforcement learning with verifiable rewards (RLVR) training. The walkthrough covers extracting final answers from boxed outputs, normalizing them, checking mathematical equivalence, and running evaluation on the MATH-500 dataset.

AISatya Nadella says any pursuit of superintelligence must help humanity and remain under human control, and that AI benefits should spread across countries, communities, and companies. He argues for a frontier ecosystem where closed and open-source models both thrive, and that organizations should keep control of their tacit knowledge and learning loops without depending on a single model provider. Microsoft plans to publish its first-party MAI models' "Code of Conduct" for public consultation tomorrow.

AIShanghai AI Lab's InternLM team released Intern-S2-397B, a multimodal foundation model for scientific intelligence and long-horizon agents. The model uses visual pre-training on raw scientific literature pages, multi-task reinforcement learning across more than 20 scientific domains, and agentic reinforcement learning in sandboxed environments.

AIMike Knoop says he sees a path to an ARC-AGI-4 benchmark focused on open-ended invention, which he calls the gating capability between zero-sum automation and positive-sum innovation. He argues that coordinated slowdown efforts would likely apply to everyone, including open-source work, and cites chain of thought and the transformer as inventions that grew out of open science research. He concludes the research frontier must stay open to keep humanity on a positive-sum path.

AIMeituan's LongCat team hosted the Hugging Face team at its Shanghai office to discuss future model development and open-source plans. The two sides also covered recent features and technical details of Hugging Face Transformers. The post frames the visit as a step toward wider collaboration across the open-source ecosystem.

AINathan Lambert has compiled a reading list of open-model writing covering why labs release open weights, the open-versus-closed debate, and US-China competition, last updated 15 September 2026. The list includes pieces on open-model economics, safety and marginal-risk research, and recent Chinese releases such as Kimi K3 and GLM-5.2. It also cites lawmaker inquiries into Western companies' use of Chinese models.

AIShanghai Artificial Intelligence Laboratory has released Atria Dawn Preview, an agentic model built on the 744B-parameter MoE GLM-5.2 foundation model, with a 256K context window. The release page reports benchmark results across search, coding, tool use, productivity, and cybersecurity, and describes text-only setup for Codex and Claude Code.

Why it matters: The release page gives a full benchmark table against named rivals and setup steps for Codex and Claude Code, useful for anyone evaluating agentic models.

AISakana AI released Fugu Max and Fugu Ultra v2, multi-agent orchestration systems that route tasks across a pool of open-weights and specialized models. The source says Fugu Max delivers performance within striking distance of elite models at two to six times lower cost, while Fugu Ultra v2 outperforms Opus 5 and Fable 5 on Chartography and outperforms models costing three to five times more per token on DeepSWE.

AIOllama says users can customize which models appear in ChatGPT's model selector through settings. Both local and cloud models can be selected for display.

AIThe ChatGPT Desktop app, referred to as the Codex app, can now be configured to run Ollama models. Users can download or update to Ollama 0.34 to get started.

AITogether AI expanded its Fine-Tuning service with support for newer open-weight models, live metrics tracking, and finer training controls. Expert LoRA adapters can be applied to Mixture-of-Experts expert layers, and early stopping keeps the checkpoint with the best validation loss. Dataset previews, sample weights, pre-flight validation, and lower prices on selected models are also included.

AIAi2 released AstaBrief-8B-SFT, an 8B intermediate supervised fine-tuning checkpoint built on Qwen3-8B that turns a research question and retrieved literature excerpts into a cited report. On the ScholarQA-CS2 test set of 100 computer science questions, it scored an average of 83.7 versus 77.3 for base Qwen3-8B, with citation recall at 71.3 versus 64.6. The model is licensed under Apache 2.0 for research and educational use.

AIModel page:

AIGiles Thomas extended the GPT-2-style code from Sebastian Raschka's "Build a Large Language Model (from Scratch)" into a 6-expert, 2-active mixture-of-experts model and trained it from scratch over 8 days. Raschka praised the project as interesting LLM work done on a single GPU.

AICognition has welcomed Jonathan Kelley and the Dioxus team, whose framework Cognition used extensively to build and improve Devin's performance. Cognition plans to continue supporting Dioxus, Blitz, Taffy, and Subsecond while increasing investment in Dioxus-Native and Blitz. The Dioxus team will also work on Devin's virtual machine, computer use skills, and testing capabilities.

AISGLang and Miles ship day-0 inference and RL support for DeepSeek V4.1 Flash, with weights now available. The model is natively multimodal with 552B backbone parameters, 16B active during decode and 8B during prefill, and supports up to 1M context. V4.1 adds shared compressed KV across layers, a two-stage sparse indexer, and a 196B Engram lookup memory.

AIFireworks AI describes a four-stage path for teams moving from renting closed frontier models to training their own, starting with API use and prompt, context, and harness engineering. The post uses the UIPad computer-use dataset to show that Kimi K3 ties GPT 5.6 Sol overall at 87.7 but wins three of four categories while costing about half as much, suggesting routing. After roughly three hours of training on the training split, the tuned Kimi K3 outperforms GPT 5.6 Sol on the held-out test set.

AIGoodfire used Ai2's open post-training stack, including the Dolci preference dataset, intermediate Olmo checkpoints, and OLMES evaluations, to trace a safety regression in Olmo. Preference training made Olmo more likely to comply with harmful requests on a refusal benchmark, and Goodfire linked part of this to specific Dolci examples where the preferred response encouraged compliance. Because Ai2 publishes the individual preferred and rejected responses, researchers could test targeted changes to reduce the regression.

AIIan Johnson introduced Latent Craft, a new way to explore large datasets with UMAP, letting users fly through and collect images from them. The demo covers all 1 million images explorable in the browser, drawn from a dataset of 1,080,814 public domain images, mostly from 19th-century books, shared on the Hugging Face Hub.

AIGoogle announced general availability of ADK for Kotlin 1.0, a Kotlin Multiplatform framework for building AI agents on servers and Android. Version 1.0 reaches feature parity with ADK 1.0 Core and adds Android extensions for on-device models, cloud Gemini via Firebase AI Logic, and persistent sessions and memory with Room and AppSearch. The post includes a server-side incident triage example using KSP-generated tools and skills, plus an Android financial assistant example with human confirmation for transfers.

Why it matters: The post names the new Android and server-side capabilities and the code setup, helping Kotlin developers judge whether ADK fits their agent projects.

AICohere Labs has released Tiny Aya Base 32K, an open-weights pretrained model with 3.35 billion parameters and a 32K context window. The model covers 70+ languages, including many lower-resourced ones, and is designed for downstream adaptation and long-context research. It is a base model that has not been instruction-tuned, and it is licensed under CC-BY-NC.