DeepSeek V3.1 price slashed; 57.7k stars added to AI projects
57,684 stars today91 new projects+3,433 archify
By olud.ai editorial · built from our own tracking data, published daily
DeepSeek slashed the input price of V3.1 by 55%, dropping it to $0.25 per 1M tokens. Other models saw both increases and decreases: DeepSeek V4 Flash rose 14%, GLM 5.2 fell 18%, and Gemini 3.7 Flash batch doubled to $0.38 per 1M tokens. The moves create a mixed pricing landscape across the ecosystem.
The tracking index gained 57,684 GitHub stars across its 10,000+ open-source AI projects today. A total of 91 new projects entered the index. The weekly count now stands at 1,877 new projects and 100 new Hugging Face spaces.
Seven emerging projects were detected today, spanning agentic tools, multimodal embeddings, and creative coding. The research community also turned attention to distilling GitHub repositories into reusable skills for AI agents.
Releases of the day
Four minor releases shipped. Activepieces 0.90.1 fixes a piece upgrade migration bug and includes AWS IAM role improvements. Pydantic AI v2.38.0 introduces a compatibility change for one-off capabilities, giving them default IDs and a combine rule for repeats. Phoenix MCP 4.3.7 updates its dependency to phoenix-client 7.8.0. Vespa v8.750.13 updates the command-line tool for Vespa.ai instances. These releases are focused on stability and compatibility rather than new features. Activepieces users get a flow migration fix. Pydantic AI developers need to review the new combine rule for one-off tasks. Phoenix and Vespa users receive dependency bumps and minor tool improvements.
Seven models saw price changes. The largest drop is DeepSeek V3.1 input down 55% to $0.25. Other decreases: GLM 5.2 input down 18% to $0.97, Qwen3.8 2.4T A95B batch input down 20% to $2. Increases: DeepSeek V4 Flash input up 14% to $0.08, DeepSeek V4 Pro input up 1% to $1.03, Nemotron 3 Ultra input up 20% to $0.6, Gemini 3.7 Flash batch input up 100% to $0.38. DeepSeek models moved in opposite directions, creating a mixed signal for its lineup.
Seven projects entered the index today. OpenExecutive (3,179 stars) is an AI-powered virtual executive team using eight specialist Claude agents. WeMM-Embedding (1,000 stars) from Tencent's WeChat Vision Team is a family of universal multimodal embedding models. DiffusionOPSD (239 stars) applies on-policy self-distillation in diffusion models. CarWatch (270 stars) turns a car into a chat-room agent using a Raspberry Pi 5 and dashcam. doop (540 stars) is an open-source alternative to Paper.design, a multiplayer design canvas with MCP. hunk (8,958 stars) is a review-first terminal diff viewer for agentic coders. ainovel-cli (1,886 stars) is a multi-agent CLI for automated novel generation in Chinese.
The Hugging Face community paper Repo-To-Skill distills GitHub repositories into reusable skills for AI agents. The authors argue that autonomous agents combining a model backbone with planning, execution, memory, and verification tools can carry out ML research end to end. Repo-To-Skill narrows the gap between general-purpose codebases and domain-specific agent capabilities.
The day's facts point to a maturing ecosystem where pricing is volatile as providers jockey for position, new projects cluster around agentic workflows and multimodal understanding, and research continues to push toward autonomous AI researchers. Users should expect further price corrections and more specialist tools emerging from the open-source pipeline.
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
