$ the-wire · showcase
GPT-6 Luna results land in Rails evals, site, and devcontainer gets Ruby 3.4.11
By RepoJournal · Filed · About Rails · Composed from the cited sources · methodology
2 people shipped this
Rails published a second-stage benchmark run of GPT-6 Luna at max effort across its agent evals and surfaced the numbers on the website, while the devcontainer picked up Ruby 3.4.11.
Agents on Rails /ai: Rails World 2026 data (+ GPT-6 Luna at max effort) rails/website
The /ai Stage 2 table and cost charts now include GPT-6 Luna at max effort, so the public Agents on Rails page reflects the same configuration the eval repo just ran. Anyone comparing agent models on the Rails World 2026 data should re-read the numbers rather than the older Stage 2 snapshot.
Add Stage 2 runs: GPT-6 Luna at max effort rails/ai-evals
Sixty runs of GPT-6 Luna at max effort on the 20 full-feature tasks were added to rails/ai-evals, giving the suite its first Stage 2 results for that configuration. The published spread is now wide enough to compare max-effort runs task by task instead of on a single score.
Add GPT-6 Luna max runs rails/ai-evals
A companion commit lands the GPT-6 Luna max run data itself in the eval repository, with Vladimir Dementyev credited as co-author. This is the raw input behind the summary added in the other pull request.
Agents on Rails /ai: Rails World 2026 data (+ GPT-6 Luna at max effort) (#743) rails/website
The website change was merged, putting the Rails World 2026 /ai data live; the merge note thanks the author for the quick turnaround. In practice the evals repo and the published page are now in sync.
Add Ruby versions: 3.4.11 [automated-ruby-update] (#138) rails/devcontainer
The automated Ruby bump rolls rails/devcontainer forward to Ruby 3.4.11, which is what new containers built from this image will run. Rebuild if you pin a container image and care about the patch level.
