Today's take
The clearest signal is a shift from agents that execute tasks toward workflows with reusable methods and inspectable results. Skills, code review and creative tools are attracting attention together, extending the focus to quality control and delivery.
The latest available OpenRouter data, dated 2026-10-09, puts personal and coding agents at the top of app usage, echoing that engineering direction. The GitHub snapshot is dated 2026-10-11, so this is a comparison of signals across dates, not evidence of same-day causation.
Open source
Today's Top 50 includes 1 new entrant, but the more informative movement is in tools that structure established workflows. Three themes stand out: reusable skills, review mechanisms and concrete creative outputs.
- Skills as engineering assets: #2 mattpocock/skills gained 1,679 stars and climbed from #5 to #2. #30 affaan-m/ECC entered the Top 50 with 276,296 total stars and +411 stars in 1 day, combining skills, memory, security and research-first development; entering the ranking does not establish a recent launch.
- Review inside the agent workflow: #10 alibaba/open-code-review jumped from #39 to #10 with +925 stars in 1 day. Its combination of deterministic checks, an LLM agent and line-level comments points to demand for inspectable generated code.
- Creative tools built around deliverables: #1 storytold/artcraft added 2,944 stars and retained the lead, targeting workflows for artists, designers and filmmakers. #4 cathrynlavery/diagram-design added 1,383 stars with 44 diagram types and HTML + SVG output, packaging visual communication as an agent capability; its move from #2 to #4 also shows that star gains and ranking direction can diverge.
Models
Model usage on October 9 shows a relatively stable top tier with selective growth. Absolute Token volume and growth rate tell different stories.
- Usage leaders: DeepSeek V4.1 Flash (batch) processed 5.48T Tokens, up 2% from the previous day. GLM 5.3 Flash (batch) recorded 1.97T, up 11%, while GPT-6 Luna (batch) recorded 1.13T, down 1%.
- Growth within the top ten: #9 GPT-6.1 Sol (batch) reached 406B Tokens, up 55%. Claude Opus 5.5 (batch) recorded 728B, up 1%, without a comparable surge.
- Movers outside the top ten: Muse Spark 1.3 reached 165B Tokens, up 120%, and Claude Sonnet 5.5 (batch) reached 311B, up 60%. These are growth signals among models ranked 11–50, not the overall usage leaders.
Apps
Personal agents and coding tools appear prominently in both usage and growth rankings. App growth is measured against the average of the preceding three equal-length periods; no numerical growth rates were supplied.
- Personal agents lead: Hermes Agent ranked #1 for both usage and growth with 1.99T Tokens. Its persistent memory and reusable skills align with the open-source emphasis on accumulating agent capabilities.
- Coding workflows remain active: Claude Code recorded 1.55T Tokens, Codex 713B, Kilo Code 634B and Cline 547B. All four also appeared in the top five growth entries, supporting the workflow theme without establishing which models powered their usage.
What to watch
Watch whether skills, review and memory tools sustain attention alongside rising agent app usage. Model and app figures cover only publicly attributed traffic through OpenRouter: Tokens measure usage, not people or ecosystem-wide market share.
