Open-source AI projects gained 57,684 GitHub stars today, with 57 new projects entering the index. This brings the weekly total to 2,167 new projects and 100 new Hugging Face spaces.
Several model prices changed today, with both increases and decreases across providers. Releases landed for Windmill, Pydantic AI, Phoenix, and Puppyone, each adding features or fixes.
Windmill v1.811.0 makes snowflake_oauth work as a dbt warehouse on every engine, expanding database integration. Pydantic AI v2.43.0 adds a first-run banner that describes the run and opens clai sessions with it. Phoenix v20.11.0 introduces GraphQL mutations for span, trace, and session notes, and serves the shared skills root on the /mcp endpoint. Puppyone v0.3.19 is a signed and notarized desktop release promoted from internal testing.
Windmill v1.811.0 feature1.811.0 (2026-09-12) · Features · * make snowflake_oauth work as a dbt warehouse on every engine (#11095) (9fc50a2) · Bug Fixes · * accept any hub version of the git sync...
Pydantic AI v2.43.0 What's Changed 🚀 Features * Add a first-run banner describing the run, and open clai sessions with it by @DouweM in https://github.com/pydantic/pydantic-ai/pull/7447 🐛 Bug...
Phoenix arize-phoenix-v20.11.0 20.11.0 (2026-09-12) · Features · * annotations: add GraphQL mutations for span, trace, and session notes (#16053) (e374a4d) * mcp: serve the shared skills root on the /mcp...
PPuppyone v0.3.19 Signed and notarized PuppyOne Desktop release. · Commit: 98f296043c034b6671db9c355f0d111957b2a3d4 Promoted Internal release: v0.3.19-internal.39 R2 version prefix:...
GLM 5.2 input price rose 114% to $0.6 per 1M tokens. DeepSeek V4 Pro input price went up 9% to $0.95, while DeepSeek V3 input price dropped 14% to $0.25. Qwen3 30B A3B Instruct input price increased 80% to $0.09. Kimi K3 input price fell 40% to $1.8. Solar Pro 4 input price surged 200% to $0.09. The changes affect users of these models through cost adjustments for input tokens.
A Hugging Face community report (191 upvotes) from the NCP Team introduces NCP-ArchPreview, a latent-space language model that learns through Next Concept Prediction alongside standard next-token prediction. The model predicts discrete concepts in addition to tokens, aiming to push autoregressive pretraining beyond token-level generation.
The day’s wide price swings signal ongoing market recalibration across model tiers, while incremental releases improve tooling and debugging. The research pick points to interest in alternative pretraining objectives that move beyond traditional token prediction.
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
← All AI news