Price cuts hit small models, n8n and Phoenix ship fixes
59,594 stars today83 new projects+3,953 dsh-our-free-model
By olud.ai editorial · built from our own tracking data, published daily
OpenAI cut gpt-oss-20b input pricing by 33 percent to $0.02 per 1M tokens, the largest drop among several price moves today. Moonshot AI lowered Kimi K2.5 input price by 6 percent to $0.5, MiniMax reduced M2.5 by 10 percent to $0.27, and Alibaba trimmed Qwen3.5-9B by 6 percent to $0.16. Z.AI cut GLM 5.2 batch input price by 40 percent to $0.32 while raising GLM 5.3 Flash batch by 25 percent to $0.05.
The ecosystem gained 59,594 GitHub stars across tracked projects today, with 83 new projects entering the index. This week saw 1,793 new projects added plus 100 new Hugging Face spaces. The top gainer, dsh-our-free-model, added 3,953 stars.
Releases of the day
n8n shipped version 2.47.7, a bug fix release that stops redacting executions using global credentials. The fix addresses issue #43776 and affects workflow execution visibility. ms-swift v4.6.1 added support for several new models including NVIDIA-Nemotron-3.5-Super-VL-120B-A12B and NVIDIA-Nemotron-3.5-Transcribe-30B-A4B, plus OCR models Unlimited-OCR-2 and MiniCPM-OCR-1.0. Phoenix v21.1.0 added a dataset-versions list CLI command and new pnpm dev scripts for UI development.
Price moves split by model size. Small models saw cuts, with gpt-oss-20b down 33 percent and Qwen3.5-9B down 6 percent. Larger models moved up, with GPT-5.6 Sol rising 10 percent to $4.4 per 1M tokens and Qwen3 14B up 20 percent to $0.12. Batch pricing diverged at Z.AI, with GLM 5.2 batch down 40 percent and GLM 5.3 Flash batch up 25 percent.
▼ 6%Kimi K2.5Moonshot AI · input $0.53 → $0.5 per 1M tokens
▼ 33%gpt-oss-20bOpenAI · input $0.03 → $0.02 per 1M tokens
Pagekin entered the index today with 2,600 stars. The TypeScript project is a free AI novel writing tool supporting setup, blueprint, drafting, review, and finalization. It offers decomposition, continuation, and rewriting features, and can import writing skills.
The research pick is SO-Arena, a benchmark for System One models. These models aim for fast, general-purpose decisions without lengthy token generation. The paper argues current benchmarks miss their capability spectrum and reliability, so SO-Arena tests beyond aggregate performance.
Today's price cuts cluster on small and mid-size models while flagship prices rise, suggesting providers are pushing cost efficiency down the stack. The batch pricing split at Z.AI points to tiered strategies for different workloads. Release activity stayed maintenance-focused, with n8n and Phoenix shipping fixes rather than new features.
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
