DeepSeek raises prices, Mistral cuts; n8n and Windmill patch
197,032 stars today59 new projects+10,459 rea
By olud.ai editorial · built from our own tracking data, published daily
DeepSeek raised input prices across several V4 models on October 30, 2026. The largest move was GLM 5.3 Flash from Z.AI, up 400% to $0.15 per 1M tokens. Mistral Large 4 dropped 49% to $0.7 per 1M tokens.
The ecosystem tracked 197,032 GitHub stars gained across 10,000+ open-source AI projects today, with 59 new projects entering the index. The top gainer "rea" added 10,459 stars. This week, 2,171 new projects entered the index plus 100 new Hugging Face spaces.
Release activity was light but steady. n8n, Windmill, Unstructured, and Phoenix all shipped minor updates. No new models were announced today.
Releases of the day
n8n released version 2.45.7 with bug fixes. The core now requires execution:read permission when workflow-scoped routes read execution data, and IMAP trigger expressions were resolved. Windmill shipped v1.854.2, fixing a bug where a new app's links were read before its session draft was discarded. Unstructured updated to 0.27.36 with maintenance changes and an XML fix. Phoenix released arize-phoenix-client v3.6.0, adding DECISION and AGENT annotation kinds and exposing prompt version creation timestamps through the client API.
Input prices moved in both directions today. DeepSeek V4 Pro 0423 rose 80% to $1.6 per 1M tokens, DeepSeek V4 Pro 0813 rose 30% to $0.13, and DeepSeek V4.1 Flash 1017 rose 300% to $0.04. Z.AI's GLM 5.3 Flash rose 400% to $0.15, while GLM 4.6 fell 9% to $0.5. Moonshot AI's Kimi K3 rose 26% to $0.24. Mistral Large 4 fell 49% to $0.7. Users of DeepSeek's newer V4 variants face the steepest per-token cost increases, while Mistral and one GLM tier became cheaper.
▲ 80%DeepSeek V4 Pro 0423DeepSeek · input $0.89 → $1.6 per 1M tokens
A new project named "awesome-freellm-apis" entered the index today with 4,895 stars. It lists 134+ free LLM APIs and AI API keys from 40+ providers, including Google Gemini, NVIDIA NIM, Groq, and OpenRouter, with one-click setup for Claude.
The Hugging Face community research pick is "Intrinsic Distribution Matching" by Deyuan Liu, Yansen Han, Kaiwen Zheng and colleagues. It addresses one-step and few-step text-to-image generation, where models must render prompt details accurately with few refinement chances. The method uses distribution matching distillation with score differences to align generated outputs.
Today's price moves show input token costs diverging sharply across providers, with DeepSeek's newer models becoming notably more expensive while Mistral undercuts on price. The steady flow of patch releases and community resource projects suggests the ecosystem is consolidating tooling and cost efficiency rather than pushing major model launches.
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
