Today's take
AI ecosystem shows clear acceleration in Agent engineering, with open-source projects making significant progress in toolchain development. This trend aligns with the growth of DeepSeek and GLM series models, while CLI agents dominate token consumption.
Open source
The open-source side shows a clear trend towards Agent engineering. morluto/rea, ranking first, provides reverse engineering capabilities that support Agent development. wheresryan22/anatomy enhances Agent's system understanding through visualization tools, while mattpocock/skills focuses on engineering skill accumulation.
#1 morluto/rea https://github.com/morluto/rea 406,391 stars, +5,749 stars in 1 day. This project provides reverse engineering capabilities for Agents, reflecting developers' deep needs for Agent toolchain support.
#32 wheresryan22/anatomy https://github.com/wheresryan22/anatomy 22,458 stars, +741 stars in 1 day. This tool enhances Agent system understanding through visualization, showing the trend of Agent engineering towards usability.
#40 mattpocock/skills https://github.com/mattpocock/skills 306,137 stars, +614 stars in 1 day. This project focuses on engineering skill accumulation, reflecting the importance of Agent capability reuse.
Models
The model side shows a clear concentration of the top players. DeepSeek V4.1 Flash 1017 leads with 6.25T token usage, while GLM 5.3 Flash (batch) shows a 31% daily growth, becoming a major breakthrough. The model growth aligns with the development of open-source Agent toolchains.
- #1 DeepSeek V4.1 Flash 1017 6.25T tokens (+11%)
- #3 GLM 5.3 Flash (batch) 2.35T tokens (+31%)
- #26 Hy4 preview 109B tokens (+190%)
- #25 Qwen3.8 Flash 110B tokens (+34%)
Apps
CLI agents dominate the application side, with Hermes Agent leading at 1.99T tokens. Codex and OpenClaw, as the fastest-growing applications, directly relate to the open-source Agent toolchain.
- #1 Hermes Agent 1.99T tokens
- #2 Claude Code 1.4T tokens
- #3 Codex 869B tokens (+13.2M requests)
- #5 OpenClaw 646B tokens
What to watch
Agent engineering is moving from technical exploration to engineering practice, with open-source toolchains, model optimization, and application landing forming a positive cycle. Future attention should focus on Agent skill reuse mechanisms and cross-platform compatibility.