A directional reading of adoption and attention across four public signals. It is not active users, market share, sentiment, or a product-quality ranking.
Benchmark snapshot · separate from the indicator
Model strength
Two decisions, two pairs: peak capability and a practical daily driver. Compare each test directly rather than treating different models as one universal score. Checked ; source results may change.
Peak capability
For the hardest work
The flagship coding choices from each provider. This is about capability, not the cheapest way to complete a routine task.
The lead changes with the task: Astra leads on terminal work here; Fable leads on this repository-fix set. Neither result establishes an overall winner.
Daily driver
For day-to-day work
Balanced models for regular coding. The compared settings are Sol medium and Sonnet high; effort labels are not identical across providers.
Codex GPT-6.1 SolClaude Code Claude Sonnet 5.5
Terminal-Bench 4.0 ↗Artificial Analysis · model-level evaluation · medium vs high effort
48%44%
SciCode ↗Artificial Analysis · model-level evaluation · medium vs high effort
53%54%
Close on these coding tests. In the same evaluation, estimated API cost per task is $0.21 for Sol and $0.88 for Sonnet; this is not subscription pricing.
These are curated source snapshots, not live scores. Model-level evaluations and agent-harness tests are labeled separately; results and relative strengths can change with settings, tasks, and benchmark versions.
Methodology
How the index is calculated
Each channel contributes its Codex share of the Codex + Claude Code total. The indicator is the weighted sum: 45% Homebrew installs, 30% new GitHub stars, 15% npm downloads, and 10% VS Code Marketplace installs. Claude Code completes the reading to 100.
The first three channels use the same completed 30-day window. VS Code reports cumulative installations, so this is a mixed-horizon indicator, not a pure 30-day momentum measure. The weights are editorial choices, not a statistical estimate of market share. A new reading requires all four complete sources; otherwise the previous reading is marked stale.
Homebrew analytics cover only one installation path; GitHub stars measure interest; npm downloads include automated builds and Claude Code's npm installation path is deprecated. Marketplace counts are not active users. Formula v2 starts a new history and is never blended with the previous two-channel 70/30 series.