Langfuse and MLflow ship updates as DeepSeek cuts prices
98,892 stars today93 new projects+4,079 Agent-Reach
By olud.ai editorial · built from our own tracking data, published daily
DeepSeek cut input prices on three models on October 7, while Z.AI raised GLM 5.2 pricing by 400 percent. The moves came as the 10,000-plus open-source AI projects tracked by the index gained 98,892 GitHub stars and added 93 new projects. This week, 1,951 new projects and 100 Hugging Face spaces entered the index.
Release activity stayed broad across observability, workflow automation, and vector database tools. Langfuse and MLflow both shipped feature releases, while several smaller tools issued maintenance updates. The weekly project inflow remained steady alongside the price changes.
Releases of the day
Langfuse v4.54.0 added trace classification into buckets and support for external media storage. MLflow 3.17.0 introduced TypeSafe models across evaluation, gateway, and tracing workflows. Weaviate v1.40.0 brought namespaces to general availability, adding control-plane and data isolation between users on a shared cluster. Browser Use 0.13.11 added toolsets for Claude, requiring an Anthropic SDK release with the needed support. Cline v4.1.23 changed behavior so that a model response without a recognized finish reason triggers one continuation request instead of being treated as complete. Open Interpreter rust-v0.0.56 added GPT-6.1 Sol metadata and refreshed its maintained provider catalog.
Input prices moved in both directions. DeepSeek cut V4 Flash 0423 by 33 percent to $0.02 per million tokens and V3.1 Terminus by 10 percent to $0.27, while raising V3.2 by 8 percent to $0.28. Alibaba cut Qwen3.5-35B-A3B by 47 percent to $0.08 and Qwen3.5 397B A17B by 18 percent to $0.45, but raised Qwen3 30B A3B Instruct 2507 by 100 percent to $0.10. Z.AI raised GLM 5.2 by 400 percent to $0.15. Moonshot AI cut Kimi K3 by 39 percent to $0.79.
▼ 39%Kimi K3Moonshot AI · input $1.29 → $0.79 per 1M tokens
Emerging projects
Agent-Reach led GitHub today with 4,079 stars. The index added 93 new projects today and 1,951 this week, along with 100 new Hugging Face spaces.
Research pick
The research pick, shared on Hugging Face with 63 upvotes, rethinks cross-tokenizer on-policy distillation. On-policy distillation trains a student on its own generations using teacher feedback. When teacher and student use different tokenizers, comparing predictions requires alignment at both sequence and vocabulary levels. The paper argues that alignment coverage alone is not enough and introduces supervision reliability as a separate concern.
Today's price moves show no single direction in inference costs, while release activity stayed broad across observability, workflow, and vector database tools. The continued inflow of new projects and GitHub stars points to steady ecosystem growth.
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
