Skip to content
RubyInsights

Lexxy 1.0 Ships, Rails Main Requires Ruby 3.3.5, and Three Frontier Labs Land on $2/$10

Rails gets a real rich text editor, main drops dynamic controller routes for good, Tenderlove answers DHH at Rails World, and Anthropic, OpenAI and Google all price their new workhorse models at $2/$10

Also available in Português (Brasil)

diarioaction-textOpenAIrailsrails-worldClaudelexxyrubybrakemangeminiia
Share

Rails

Lexxy 1.0 is out, and it is coming for Trix. Jorge Manrubia announced Lexxy 1.0 on September 28. It is built on Meta's Lexical, powers Basecamp 5, and brings what Trix's flat block model never could: tables, Markdown shortcuts, syntax-highlighted code blocks, image galleries, mention/emoji prompts and attachment previews. It is Action Text native, emitting the same markup Action Text already stores, so switching is one line with Rails 8.2:

config.action_text.editor = :lexxy

37signals' guidance is blunt: new apps should use Lexxy, and standard Action Text setups should switch. They also plan to make it the Rails default. Worth noting for anyone building with agents: the team leaned on AI agents plus Playwright to write over 600 browser tests.

Rails main: three changes to watch for in 8.2.

  • #58908, merged by byroot on Oct 1: minimum Ruby is now 3.3.5. Earlier 3.3 patches carry a WeakMap use-after-free; requiring 3.3.5 lets Rails drop its WeakKeyMap polyfill and stop iterating connection pools unnecessarily for the query cache. If you are pinned to an older 3.3, plan the bump before 8.2.
  • #58893, merged by tenderlove on Sept 30: dynamic :controller / :action route segments now raise ArgumentError. Deprecated since Rails 5.0. Grep your routes.rb for ":controller(/:action(/:id))" now, not on upgrade day.
  • #58876, merged Sept 27: except_on is now treated as a callback option, so errors.added? matches those validations and errors.details stops leaking the option.

The Sept 25 This Week in Rails also flagged a rewritten Asset Pipeline guide focused on Propshaft and Ractor-friendly connection handling in Active Record (#58306).

Rails World talks are already on YouTube. One week after Austin, all recordings are online. Two to prioritize:

  • Aaron Patterson's closing keynote (video) is the counterweight to DHH's "pencils down" opener. His core argument: compilers give you the As-If Rule, a contract that observable behavior matches your code, and "your AI has not made this promise to you." He backed it with ZJIT and Ractor work, including Ractor-local GC making allocations about 70% faster, and used the May RubyGems incident as a case study in unreviewed automated code. (Global Nerdy's writeup)
  • Matz and DHH's fireside chat (video) covers Ruby's role in an AI-heavy world, Spinel, and DHH reaffirming Rails for new web apps.

Gem releases

  • Solid Cable 4.1 (Sept 26, "Rails World Edition"): encrypted messages, consolidated writes, and less hashing during polling. If you run Action Cable on the database, this is a free win.
  • Brakeman 8.1 (Sept 30): Sonar generic issues format, regex shown in validation warnings (fixing duplicate fingerprints), the validation check extended to non-Active Record models, positional enum syntax in SQL injection checks, and faster vendor skipping. Update it in CI.

AI

The $2/$10 week. Three labs shipped workhorse models at the same headline API price, $2 per million input tokens and $10 per million output:

  • Claude Sonnet 5.5 (Sept 28, claude-sonnet-5-5): Anthropic says it is 30%+ faster than Sonnet 5 and up to 30% cheaper per task. It posts 70.6% on Terminal-Bench 4.0 and 55.5% on CursorBench 4.0, close to Opus 5.5's 57.8%.
  • GPT-6.1 Sol (Sept 29, gpt-6.1-sol): pitched as near-Astra intelligence at a fifth of the price, matching GPT-6 Astra on DeepSWE 1.1. Cached input is $0.10.
  • Gemini 4 Argon (Sept 30): 77.9% on DeepSWE v1.1 and a 1M-token output limit, but you can't use it yet. It is rolling out first to vetted cyber defenders through the Fairwind Program, without cyber guardrails for that group. The $2/$10 rate is introductory and rises to $4/$20.

For Rails teams running RubyLLM or similar, price parity means model choice now comes down to your own evals and latency, not the rate card. Swapping providers behind one interface has never been cheaper to test.

OpenAI DevDay. The recap includes computer use in the Agents API, always-on "dots" agents, an Ultrafast mode (up to 6x faster in the API), Bedrock Managed Agents with AWS, and a new Pro 500 consumer tier.

OpenAI held back GPT-6.1 Astra. Per CBS News (Sept 28), head of safety systems Saachi Jain said the model "didn't quite meet the bar in terms of staying within scope and authorization." Agent scope control is now a release blocker at the frontier, and it should be in your app too.

Distillation, disclosed. OpenAI published details (Sept 30) of a July campaign, linked to individuals associated with Moonshot AI, that tried to extract hidden reasoning. At its peak it sent 16,000 requests from more than 4,000 accounts. OpenAI shut it down by July 28.

Also: Anthropic committed $100M (Oct 2) to train 10,000 "Frontier Deployed Engineers" by the end of 2027, through a residency-style program.

Sources

Comments

Sign in with Google or GitHub to comment.