Social Wire
Page 5 of 49 · newest first.
@cognition · Cognition
Gemini 4 Carbon achieves near state-of-the-art scores on FrontierCode 2.0, a coding benchmark that tests a model's ability to go beyond the spec, with a score of 40% and a cost of $4.04 per task.@david_zhang_sf Ambassador · Cognition
David Zhang is organizing a Devin meetup in Ho Chi Minh City next week (Sat 7 Nov, Joi Cafe, with AGENTIQ.CLUB) — a short Devin intro plus coworking. luma.com/pg02c4z6@dzienko Ambassador · Cognition
dzienko: Cognition buying Corridor means coding agents get their own security layer — more agent-written code means more need to check it before merge.@Fei2411 Ambassador · Cognition
Fei2411 on Cognition's FrontierCode v2: Gemini 4 Carbon at 40.0, behind Claude Opus 5.5 (45.5) and close to Claude Sonnet 5.5 (41.0). He calls that a strong showing and flags another Google model still to come.@Fei2411 Ambassador · Cognition
Fei2411 compared GitHub Copilot Fusion (near Opus 5.5, about 30% cheaper) to Devin Fusion. He was skeptical of Fusion until Devin's; he notes the Copilot chart uses 6 Luna, while Devin users can swap in free SWE-2 for lower cost and similar performance.@Fumiya_Kume Ambassador · Cognition
Fumiya Kume: SWE-2 is free through the end of November; with a clear goal and problem it keeps running autonomously.@harshil1712 Ambassador · Cloud
harshil1712: Devin on cloud enabled Cloudflare Workers Preview from the docs, opened a PR, then clicked through the preview in a browser and sent him the recording.@kinopee_ai Ambassador · Cognition
After Devin Security 2.0 with Mythos, kinopee reads Cognition's Corridor acquisition as a clear signal they are doubling down on security tools for AI coding agents.@kinopee_ai Ambassador · Cloud
kinopee: Devin needs well-defined tasks and done criteria. He built "Devin man" on Grok Bot to organize tasks, launch Devin sessions, and babysit PRs.@kinopee_ai Ambassador · Cognition
kinopee: Devin Security Swarm also checks reproducibility on a VM — reviewing the diff alone cannot tell you true urgency or severity.@kinopee_ai Ambassador · Cloud
kinopee is running Devin heavily with Sign in with ChatGPT and says his Devin quota barely drops — unused ChatGPT/Codex capacity is covering the sessions.@kinopee_ai Ambassador · Cloud
kinopee's other cheap path: Sign in with ChatGPT — if you aren't burning Codex limits, unused ChatGPT subscription capacity can run Devin.@MatheusPagani77 Ambassador · Cognition
Matheus Pagani lists Hack Brasil prizes including $20k in Cognition credits, plus Firecrawl, Resend, and visa fast-tracks — coupons for everyone, final at Google Brasil.@kinopee_ai Ambassador · Cognition
kinopee: Devin Discord announced the SWE-2 unlimited campaign is extended to the end of November — start the promo now if you already have a code.@MatheusPagani77 Ambassador · Cloud
Matheus Pagani: Devin wrote Terraform and deployed infra on OCI live at Oracle Brasil in São Paulo. SWE-2 unlimited on Devin Pro/Max (Desktop and CLI) through 30 Nov covers Hack Brasil's online phase.@MatheusPagani77 Ambassador · Cognition
Matheus Pagani posted an X live titled "Dominando Agentic Coding com o Devin" (Mastering Agentic Coding with Devin).@sammwy Ambassador · Cognition
sammwy: SWE-2 is free through Dec 1 (Gemini 4 Argon 80% off too). He uses SWE-2 as his daily driver plus Opus and GPT Sol and still sits around 60% of weekly usage — hours of SWE-2 without the weekly meter dropping.@ScarletKc Ambassador · Cognition
ScarletKc: Devin already has an iOS app in TestFlight beta — not on the App Store yet. Apply at https://devin.ai/mobile; the invite arrives by email. He got his invite today and installed it so he can kick off coding from his phone.@willebrew Ambassador · Cognition
willebrew on the Corridor deal: Cognition showing dedication to world-class security tools for Devin.@cognition · Cognition
Cognition introduces FrontierCode 2.0, a new coding benchmark that measures a model's ability to go beyond the spec and demonstrate judgment.@Ariel_Cardenas0 Ambassador · Cloud
Ariel Cardenas built a free mystery game with Devin: casa-montenegro-one.vercel.app.@brandon_galang · Cognition
FrontierCode 2.0 is pushing the boundaries of coding benchmarks, emphasizing sensible eval criteria.@bryllim_ Ambassador · Cognition
bryllim gathered Cognition Devin users in Manila.@Chris_Wozniczek Ambassador · Cognition
Chris Wozniczek: FrontierCode is now on Artificial Analysis. He lists what it tries to measure beyond spec-following: functional correctness, expanding an underspecified prompt, and design/scoping for an existing codebase.