Price volatility hits model ecosystem as GLM 5.2 skyrockets 614%
270,384 stars today75 new projects+7,887 skills
By olud.ai editorial · built from our own tracking data, published daily
GLM 5.2 input price surged 614% to $0.50 per 1M tokens on August 11, the largest single-day price increase tracked by the system. The ecosystem gained 270,384 GitHub stars across 10,000+ tracked open-source AI projects today, with 75 new projects entering the index. Weekly context shows 1,993 new projects and 100 new Hugging Face spaces added this week.
The "skills" project leads GitHub today with plus 7,887 stars. Price moves were mixed, with four models dropping and two rising sharply. Alibaba's Qwen3.5 397B fell 22% to $0.39, while NVIDIA's Nemotron 3 Super dropped 73% to $0.08 per 1M tokens.
Releases of the day
Cline v4.1.8 shipped, letting users enter any Vertex model ID by hand, including models the catalog does not list yet, and added support for Fable 5 on Vertex. Ray 2.57.0 enabled DataSourceV2 by default, changing how read_parquet and similar functions handle scanning and listing. Langfuse v4.7.1 removed dead foreground runtime support for the in-app agent and fixed worker bounding. Pydantic AI v2.27.1 restored tool spans for failed argument validation. Phoenix released arize-phoenix-client-v3.0.0 with breaking changes that replace the google-generativeai formatter with google-genai. Composio patch release updated its json-schema-to-zod dependency to version 0.3.0. Llama.cpp b10357 focused on OpenCL optimizations for FA prefill kernels.
GLM 5.2 input price increased 614% to $0.50 per million tokens from Z.AI. Gemma 4 26B A4B input price rose 71% to $0.12 per million tokens from Google. Four prices dropped: Qwen3.5 397B fell 22% to $0.39, Nemotron 3 Super fell 73% to $0.08, Qwen3 14B fell 48% to $0.12, and DeepSeek V4 Flash 0731 fell 11% to $0.08 per million tokens.
Eight projects detected today. skill-recorder, a TypeScript desktop app with 2,805 stars, records on-screen work and reconstructs it as intent plus steps via GitHub Copilot CLI. DEEIX-Chat, at 1,262 stars, offers an enterprise AI workspace for model routing and multimodal chat. paritok-4b-v1, with 1,016 stars, is a compression gateway that cuts token bills. openloomi, at 1,008 stars, is an open-source AI coworker. airecon (930 stars) is an autonomous cybersecurity agent. Compartment (735 stars) provides encrypted offline agentic memory with a GUI. NexusRAG (421 stars) is a hybrid RAG system. pr-af (452 stars) ranks as the number one open-source code reviewer on Code-Review-Bench.
A Hugging Face community paper, SWE-Bench ProMax, benchmarks AI coding agents on large-scale multilingual code refactoring. The paper notes that existing benchmarks are rapidly saturating and a recent audit found nearly 60% of unso.
The sharp GLM 5.2 price increase against broad price drops from major providers signals a bifurcation in pricing strategy, likely reflecting supply constraints or repositioning. The emergence of eight new projects concentrated on agentic workflows, memory, and compression suggests the ecosystem continues to prioritize practical engineering tools over raw model capability.
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
