Patch releases, price hikes, and 81k stars mark the day
81,622 stars today72 new projects+30,731 rea
By olud.ai editorial · built from our own tracking data, published daily
Open-source AI projects gained 81,622 GitHub stars across the tracked index today. 72 new projects entered the index. The leading project "rea" added +30,731 stars.
This week, 1,740 new projects entered the index. 100 new Hugging Face spaces were also added. The ecosystem continues to expand steadily.
Several minor releases landed today across agent tools and automation platforms. Input prices moved sharply for multiple models. Emerging projects in agent orchestration and local-first tooling were detected.
Releases of the day
OpenClaw v2026.10.2 completes the first upgrade from stable 2026.10.1 cleanly. Successful in-place updates no longer end in errors. Hermes Agent v0.21.7 rolls up roughly 204 PRs merged since v0.21.6 into a stable desktop release for macOS. n8n 2.42.7 fixes concurrent credential tests sharing one node type registry. It also resumes tool calls for deferred tools. Docling v2.138.0 reads pictures placed on a Numbers sheet and fixes textboxes in table cells. The patch releases focus on stability and small feature gaps.
Input prices moved up for several models. DeepSeek V4 Flash 0731 rose 100% to $0.02 per 1M tokens. Kimi K2.6 rose 51% to $0.65. GLM 5.3 rose 75% to $0.07. Nemotron 3 Ultra rose 20% to $0.6. Qwen3.8 27B fell 91% to $0.04. The changes affect developers paying per 1M input tokens.
▲ 51%Kimi K2.6Moonshot AI · input $0.43 → $0.65 per 1M tokens
Four projects were detected today. openharness leads with 1,226 stars, positioning as a command center for coding agents and machines. groma.md (194 stars) maps architecture as OKF Markdown in Git. ai-runtime-lab (181 stars) focuses on deterministic systems around non-deterministic LLMs. BeatDesign (143 stars) is a local-first AI canvas and timeline for MCP agents.
The research pick is DiffusionART, with 96 upvotes on Hugging Face. It uses online reinforcement learning to fine-tune diffusion models with black-box rewards. Reverse-process methods like Flow-GRPO suffer from poor credit assignment. Forward-process methods like DiffusionNFT have other limits. DiffusionART aims to be efficient, stable, and scalable for this RL fine-tuning.
The day pairs steady patch releases with sharp price swings and fresh agent-centric projects. Price volatility suggests ongoing competition for inference workloads. The star surge for "rea" points to continued interest in agent tooling.
Source: olud.ai tracking of 10,000+ open-source AI projects, 300+ models and live provider pricing. Figures are measured, not estimated. All releases · Live pricing · Latest in AI
