SGLang v0.5.19 ships with 786 PRs, prices shift across models
57,684 stars today77 new projects+3,433 archify
By olud.ai editorial · built from our own tracking data, published daily
SGLang v0.5.19 arrived today with 786 pull requests from 214 contributors, adding new model support. The release is a minor bump but reflects sustained community activity around the inference engine.
The ecosystem gained 57,684 GitHub stars across tracked open-source AI projects so far today, with 77 new projects entering the index. This week 2,043 new projects joined the index alongside 100 new Hugging Face spaces. The top gainer today is "archify" with 3,433 stars.
Pydantic AI also shipped v2.40.0, adding a background price update function and session barge-in handling. The release continues the library's focus on agent orchestration tooling.
DeepSeek adjusted prices on multiple models today. DeepSeek V4 Flash 0423 input price rose 12% to $0.09 per 1M tokens, while DeepSeek V4 Pro 0423 input dropped 4% to $0.98 per 1M tokens. DeepSeek V4 Pro 0813 input fell 12% to $0.58 per 1M tokens. Alibaba cut Qwen3.5-35B-A3B input by 68% to $0.08 per 1M tokens and Qwen3.6 27B input by 50% to $0.30 per 1M tokens. NVIDIA raised Nemotron 3 Ultra input 5% to $0.63 per 1M tokens. Undi95 lowered ReMM SLERP 13B input 22% to $0.35 per 1M tokens.
▼ 12%DeepSeek V4 Pro 0813DeepSeek · input $0.66 → $0.58 per 1M tokens
Emerging projects
Four projects were detected today. "distilly" leads with 24,207 stars, offering a tool to distill agent reasoning into reusable skills. "a-stock-data" at 9,400 stars provides a full-stack China A-share data toolkit with 54 endpoints across 19 sources. "social-media-research-skills" at 1,930 stars adds AI agent skills for social media analysis across TikTok and other platforms. "academic-humanizer" at 1,279 stars strips AI writing tells from academic papers while preserving scholarly voice.
A Hugging Face community paper with 271 upvotes presents "Compile by Training," which turns natural-language specifications into local neural functions. The approach aims to replace repeated calls to large remote models for simple text functions, reducing cost, latency, and provider dependency.
Today's facts show a market adjusting prices aggressively on Chinese models while NVIDIA moves in the opposite direction. The SGLang and Pydantic AI releases continue incremental improvements in inference and agent tooling, while emerging projects focus on practical agent skills and data toolkits.
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
