An NVIDIA research paper presents VERA, a system that converts benchmark trajectories into over 9,000 restartable sandboxes with rubric scoring while co-evolving agent harnesses alongside model weights. The team open-sourced the environment corpus, reporting that a 27B agent scored 71.6 on AutoCoWorkBench to top Claude Opus 4.8.
Scale AI has open-sourced AgentEnv, the foundational runtime used internally to power its reinforcement learning environments for training AI agents.
GeoGuess Bench evaluates AI models on 210 worldwide geolocation photo tasks. Claude Opus 5.5 ranked first overall, while open models delivered competitive accuracy at lower inference costs.
Ramine Roane says AMD is currently the leading company contributing code to vLLM, with 502 core commits in 90 days, 13% of all organization contributions, and nearly four times Nvidia. He also credited Red Hat, IBM, Embedded LLM, and Inferact.
Zixuan Li says GLM-5.3 has helped defend 389 open-source projects and found 4,249 potential vulnerabilities so far. OpenVuln remains free, and findings go privately to maintainers.
The OpenClaw Foundation is open-sourcing OpenClaw Enterprise, a control plane for persistent agents, in collaboration with Red Hat, NVIDIA, and OpenAI. It is built to run on an organization’s own infrastructure and will always be free for an organization to use.
clem says NVIDIA’s acquisition lets Hugging Face hire people it could not as a small startup and give them a decade to make open-source AI win. He invited those people to message him directly.
Oliver Prompts says researchers built PageIndex, a free and open-source RAG method that uses a tree index instead of embeddings, chunking, similarity search, or a vector database, and that it scored 98.7% on FinanceBench.
Qwen3.8-Omni-Flash is trained for long-horizon agent tasks across text, audio, and video, using the Qwen3.8-Next sparse mixture-of-experts design and a one-million-token context window. Two open-source frameworks, Qwen-MM-Plugins and Qwen-Live-Harness, add multimodal and real-time agent support.
SmolDataEnvs is an open-source set of more than 5,000 verifiable reinforcement-learning environment tasks for hill-climbing small models in code and data science. The release includes environments, evaluations, and training.
Token Meter is a free, open-source menu bar tool from Splunk that shows coding agent costs as a run is underway.
inclusionAI has open-sourced the Ming-Image-0.1-Design family under the MIT License: two complementary 6B models, Design and Layer, for visual-design workflows. ModelScope says Design ranks first among open-weight models on the Artificial Analysis UI/UX Design leaderboard, and that Layer leads all 12 evaluated Crello settings while running 4.3× faster than the evaluated 20B open-weight baseline.
Arena.ai places Qwen-Image-2.1 first among open-source models in both arenas. On Image Edit, it scores 1,367 points, ranks 16th overall, and trails GPT-Image-1.5-high-fidelity by three points.
On September 22, 2026, Ahmad Awais said he was launching Jev on Command Code for free, limited to headless mode or the Provider API. He also open-sourced the mod cmd-mod-jev-nudge.
Jev can be used only in headless mode or through the Provider API. Ahmad Awais also open-sourced the mod cmd-mod-jev-nudge and says the free access is for keeping an agent running while more work remains.
StepFun says onPanda, its internal tool for LLM data annotation and model inspection, is now open source. It reports a 52% lower median annotation time than manual post-editing and on-policy fidelity of ΔPPL under 1% versus the model’s resampling baseline.
Contributors from IBM, Meta, and Hugging Face present hardware-agnostic layers intended to balance frontier performance with portability for users running diverse models on diverse hardware.
StepFun has released the Step Code v0.1.0 command-line tool as open source under the MIT License, including one-command static site publishing with StepPage. StepFun reports 80.9% on Terminal-Bench 2.1 and 73.3% on Multi-Frame, its 150-task long-horizon benchmark.
Step Code v0.1.0 is an MIT-licensed CLI agent that reads and edits code, runs tests, and ships from one command line. StepFun reports 80.9% on Terminal-Bench 2.1 and 73.3% on Multi-Frame, its 150-task benchmark, and includes one-command publishing with StepPage.
The desk device shows idle, working, needs-a-decision, and done. Wahlin used Copilot CLI to create several animations and a web-based character studio, and published the project as esp32-agent-companion.