#Open-source ecosystem

AIMirroS has released AgentGarten, an open framework where AI agents act, observe and learn in simulated worlds. Code handles physics and game rules, while a neural renderer turns geometry sketches into realistic frames at more than 30 frames per second at 480p on a single GPU. Across five worlds, agents improved over rounds using a playbook of lessons, such as a companion-dog score rising from 13 to 19. The team describes it as a research environment, not a deployed robot product.

AIRadixArk says Miles runs reinforcement learning end to end on NVIDIA Vera Rubin, using SGLang rollouts, Megatron training and one container image. Agentic RL runs 64 concurrent sandboxes on the Vera CPU next to the GPUs. The linked SGLang post reports that Kimi K3 inference gained up to 20% faster FP8 MLA at batch 1 with 128K context, and a 5.9% end-to-end speedup from MoE tail fusion.

Why it matters: The post gives concrete speedup figures and a specific RL setup on early-access Rubin hardware, useful for engineers comparing inference and training stacks.

AISGLang optimized attention, MoE, and speculative verification kernels for Kimi K3 NVFP4 on early-access NVIDIA Vera Rubin hardware. Reported gains include up to 20% faster FP8 MLA at batch 1 and 128K context, 20% faster KDA verification with bitwise-identical output, and a 5.9% end-to-end speedup from MoE tail fusion. Miles, from RadixArk, uses SGLang for rollouts in end-to-end RL training on Rubin, including agentic RL with 64 concurrent sandboxes on the Vera CPU.

AIInferact, working with NVIDIA, led performance tuning and integrated the MiniMax Sparse Attention kernel. The team uses the CUDA locality domain feature on Rubin GPUs so each SM reads MoE weights from its own locality domain, making MoE layers up to 1.2x faster in decode.

AIvLLM says users can try its Vera Rubin support today with the vllm/vllm-openai:cu134-nightly image, which runs DeepSeek, Kimi, GLM, and MiniMax models. The project thanked contributors and said it expects further improvements for vLLM on Vera Rubin.

AIvLLM now supports NVIDIA Vera Rubin NVL72, with daily container builds and support for models from DeepSeek, Moonshot AI, Z.ai, and MiniMax. In early AgentX benchmarks, vLLM running MiniMax M3 delivered up to 7.84x the throughput per GPU of GB200 NVL72 at matched interactivity. The post is an early look, and the team expects further gains from ongoing optimizations.

Why it matters: The post gives specific hardware specs, kernel configurations, and benchmark figures for running vLLM on Vera Rubin NVL72, useful for teams planning deployments.

AISGLang says restructured MLA decode kernels on Rubin, which fit a deeper pipeline in 327 KiB of shared memory, deliver a 16% speedup at batch 16 with 128K context and bit-identical output. The post also reports 20% faster full FP8 MLA at batch 1 and 20% faster KDA verify kernels after keeping weights in registers and reducing synchronization. It additionally covers fusing MoE finalization, the shared expert, 8-GPU all-reduce, and RMSNorm into one collective kernel.

AIA GitHub post announces that kernel kits for DeepSeek's TP=2 and TP=4 recipes are now live. The post links to a repository named deepseek-v4.1-tensorfold-tp2-2xgb10 and gives no further details on performance or features.

AIFrancois Chollet says ARC-AGI-2 scores on Kaggle are now very good. The ARC Prize 2026 high score is 88.06%, set by Tufa Labs, and teams scoring over 85% share a $150K bonus prize.

AIOpenBMB reports that its MiniCPM5-2B model runs at 37 tokens per second on an iPhone Air. The post presents this as evidence that small open multimodal models can run on mobile devices without a cloud GPU, with NobodyWho noting the model is available in its Chat app.

AIUnderdog has released Saluki 27B under Apache 2.0, a 2-bit GGUF of Qwen3.8-27B that fits in 7.89 GB, versus 54 GB for the full BF16 model. On Underdog Bench, Saluki scores 88 against 84 for the full model, and it raises parallel tool-call accuracy to 42 from 35. It runs on stock llama.cpp, but math and reasoning drop sharply, with AIME 2025 at 79.2 versus 96.7.

AIThe vLLM Semantic Router team has released Decision 2.0, which answers multiple questions about one input in a single forward pass and outputs per-option probabilities. The post presents this as useful for routing and classification. A quoted post from Xunzhuo Liu says Decision 2.0 includes six open decision models ranging from 0.6B to 27B parameters, each topping same-size open models on the Jev Decision Index 0.3.

AITeknium announced that TinyFish, a new browser backend, is now available on the Hermes plugins catalog. According to a related post, TinyFish is added as a first-party Hermes plugin that lets agents search and fetch the live web for free, and it can be installed with hermes plugins install tinyfish.

AIOpenCode is offering free access to its Step 5 Preview for one week. The announcement highlights a 1M context window, multimodal support, and zero data retention.

AIUnsloth Studio binds remote-code approval to a fingerprint of the scanned code, so changed code requires fresh consent before it runs. It also blocks weight files that Hugging Face has flagged for malware in the path the selected loader would deserialize. The article describes these checks as one layer among several, alongside package-content scans and OS sandboxes, and notes that the scanner is not a sandbox and cannot catch every evasion.

AIInferact, working with DeepSeek, NVIDIA, and SemiAnalysis alongside the vLLM community, says joint work across models, custom kernels, and engine serving cuts time to first token (TTFT) by nearly 70% at ~100K throughput. vLLM is the open-source inference engine, and Inferact optimizes it for enterprise production deployments.

AIDeveloper Brandon Thomas's Artcraft has launched seven open source apps in Rust that aim to replicate the interfaces and tools of Adobe Photoshop, Illustrator, Premiere, Lightroom, After Effects, InDesign, and Acrobat Pro. Thomas said he used Anthropic's Claude Opus 5.5 to build the clean-room replacements, with WebAssembly versions available for browser use. The apps remain in a "super early alpha" state, and commenters have pointed out many current shortcomings.

AITogether with DeepSeek's model and kernels and NVIDIA's collaboration, vLLM reports that time to first token (TTFT) drops nearly 70% at roughly 100K throughput. The post credits Inferact and the vLLM community for the work and thanks SemiAnalysis for AgentX.

AIDevelopers are running Hermes on devices such as LilyGO watches, AIPI Lite, desk gadgets, and old Android phones after the Hermes Gadget open SDK and demo were released three days ago. The post credits @NousResearch and says more boards are landing on main through contributor PRs, with the SDK available on GitHub.

AISystem76's COSMIC desktop now requires contributors to declare no LLM-generated content in pull requests, including code, comments, and descriptions. GNOME Calendar and GNOME Extensions also restrict AI-generated contributions, while GNOME developer Michael Catanzaro argues the project should accept AI-generated bug reports. Catanzaro's case rests on memory-unsafe languages such as C, C++, and Vala, and he has shortened GNOME Security's disclosure deadline from 90 days to 30, effective August 1.

AITeknium says a plugin from @altryne's current project should be added to the plugin catalog. The post is a brief endorsement and does not describe the plugin's functions. Background from @tonysimons_ says Hermes is getting a local video editor for editing user footage with 42 FFmpeg scripts and no cloud or API key required.

AIEpoch AI reports that GPT-6 Astra scored 100% on the original EBR-bench by exploiting a card that bypasses the game's time-constraint expectations, so Epoch has banned that card from the default setting. Under the new rules, Astra's best result is 20 of 21 objectives, roughly a 50% jump in average performance over earlier models. Epoch will report revised scores only for Claude Fable 5.1, Claude Opus 5, GPT-5.6 Sol, GPT-6 Astra, and future models.

AIAn unnamed developer has open-sourced OpenWork, a tool letting AI agents work directly on files on a user's computer. It supports cloud or local models and existing skills, plugins, and MCP servers.