Today's take
The strongest signal is a more complete agent workflow: model access, deliverables, and execution audits are attracting attention together. Niko1221/Strata takes #1, while tt-a1i/archify and calesthio/OpenMontage climb sharply, pointing toward tools that help assistants complete concrete tasks.
Usage shows a related engineering pattern: coding agents occupy several leading positions, and 5 of the top ten model entries carry a batch label. The dates differ: open source covers 2026-10-06, while models and apps cover 2026-10-04; usage represents only traffic routed through OpenRouter with public attribution, and Token counts measure consumption.
Open source
The Top 50 includes 1 new entry and 3 projects classified as rising. Three themes explain the most meaningful moves: lower access friction, specialized deliverables, and agent quality verification.
- Model access moves closer to the workflow: #1 Niko1221/Strata adds 3,433 stars in 1 day, reaching 12,530 and moving from #2 to #1. Its consumer-hardware inference engine and compatible local APIs offer a route for existing tools to use local models.
- #12 Ebony-Vinyl/dsh-our-free-model: A new Top 50 entry with 695 stars added in 1 day and 1,551 total. Its description emphasizes access without login or API keys, signaling interest in reducing setup friction; unlimited free access remains the project's claim.
- Deliverables become skill products: #7 tt-a1i/archify rises from #17 with 1,032 stars added in 1 day, turning ideas, plans, and codebases into interactive diagrams. #8 calesthio/OpenMontage rises from #39 with 894 added stars and packages video production into 12 pipelines; both organize agent capabilities around inspectable outputs.
- #14 ifixai-ai/iFixAi: Up from #29, with 652 stars added in 1 day and 20,732 total. Its focus on independently auditing whether agents perform their intended tasks extends attention from generation to verification.
Models
Absolute consumption and growth tell different stories: a few entries lead volume, while free models outside the top ten show rapid gains. Batch labels are prominent, but they do not establish which workloads produced the traffic.
- Leaders diverge: stealth/space-bunny-alpha records 6.73T Token, +12% versus the previous day; DeepSeek V4.1 Flash (batch) records 3.29T, -13%. GLM 5.3 (batch) records 307B, -37%, showing that leading entries are not moving together.
- Free models post strong gains: Qwen3.8 27B (free) reaches 101B Token, +85%; Nemotron 3.5 Lightning (free) reaches 104B, +35%. This aligns directionally with interest in easier model access, without establishing a shared traffic source.
- Growth rate differs from scale: Grok 4.7 grows +113% to 32.9B Token, while GPT-6 Luna (batch) grows +13% to 649B. The former grows faster; the latter processes much more volume.
Apps
Agents define the leading usage pattern, with 5 top-ten entries explicitly classified as CLI agents. Coding tools also feature in the rising list, echoing open-source interest in packaged workflows and execution audits.
- A persistent personal agent leads: Hermes Agent records 1.99T Token. Kilo Code follows at 1.21T, Claude Code at 1.13T, and Cline at 1.04T, giving coding workflows several high-volume positions.
- Execution tools lead the rising list: omp, Freebuff, and Kilo Code rank first through third, with 801B, 815B, and 1.21T Token respectively. Rising compares usage with the average of the preceding three equal-length periods; exact growth percentages were not supplied.
What to watch
Watch whether access, delivery, and audit tools sustain attention alongside growing agent app consumption; the evidence supports an engineering trend, but does not establish causal links between particular apps and models.
