Today's take
Agents are moving beyond code generation toward completing work: local inference, web access, engineering skills and specialist tools are attracting attention together. On October 7, Niko1221/Strata leads daily star gains with +2,332, while nykooi1/vibe-wise climbs from #37 to #4—signals of demand for both execution infrastructure and better development methods.
The latest model and app snapshots cover October 5. Coding agents occupy several leading app positions, while multiple batch model entries are growing. These patterns are compatible, but there is no app-to-model breakdown to establish where individual model gains originated.
Open source
Three themes stand out across the Top 50: execution infrastructure, engineering skills and specialist task tools. Stars and ranking movements measure attention; they do not establish adoption or task performance.
- Execution infrastructure: #1 Niko1221/Strata adds 2,332 stars to reach 14,862, focusing on local inference and compatible APIs on consumer hardware. #2 lexmount/moli adds 1,765 to reach 10,293 with a headless browser for agents. They address complementary layers: running models and interacting with the web.
- Engineering skills and human understanding: #4 nykooi1/vibe-wise adds 854 stars and jumps from #37, helping people learn while AI writes code. #26 obra/superpowers adds 407 and rises from #39, combining agent skills with a development methodology. The workflow itself is becoming part of the product.
- Specialist tools join the workflow: #14 earthtojake/text-to-cad adds 582 stars and rises from #28, bringing CAD capabilities to agents. #35 Ryze-AI-Adgent/open-seo-mcp-skills adds 328 and rises from #49, connecting SEO tools and skills. General assistants are extending into tasks with concrete deliverables.
Models
Batch entries feature prominently among both usage leaders and fast growers. That makes batch processing an important traffic pattern in this snapshot, without establishing its cost, latency or output quality.
- Diverging leaders: DeepSeek V4.1 Flash (batch) ranks first with 6.25T tokens, up 90% versus the previous day. stealth/space-bunny-alpha records 4.2T, down 38%.
- Growth across GLM entries: GLM 5.3 Flash (batch) processes 1.56T tokens, up 36%; GLM 5.3 (batch) records 456B, up 49%. Both entries are expanding.
- Momentum outside the top ten: #14 GPT-5.6 Sol (batch) records 224B tokens and +143%, the fastest growth among the supplied movers. #21 GPT-6 Sol (batch), at 126B, and #23 Claude Sonnet 5 (batch), at 112B, both grow 78%. Growth rate and absolute scale tell different stories.
Apps
The clearest app signal is substantial agent usage across persistent and coding workflows. These figures cover only publicly attributed traffic through OpenRouter; model and app tokens measure usage, not people or market share.
- A personal agent leads: Hermes Agent records 1.86T tokens. Its memory and reusable-skills positioning echoes the open-source engineering theme, but NousResearch/hermes-agent ranks #49 with +242 stars today—a useful reminder that attention and usage are different measures.
- Coding workflows occupy leading positions: Claude Code records 1.06T tokens, Kilo Code 968B and Cline 914B. Five of the ten most-used apps are explicitly classified as CLI agents, alongside an IDE extension.
- Tool-oriented apps feature in growth: Strix leads the rising list with 96.6B tokens; pi records 462B at #2 and omp 676B at #4. This ranking compares usage with the average of the preceding three equal-length periods; exact growth percentages are not supplied.
What to watch
Watch whether attention to execution tools and engineering skills is accompanied by sustained agent usage: the directional connection is visible, while specific routing relationships and practical outcomes still need evidence.
