Is Cursor's new model actually good? Testing New UI + Composer 2

Описание: Cursor just dropped Composer 2, their in-house coding model trained exclusively on code. It's 86% cheaper than Composer 1.5 and claims to beat Opus 4.6 on coding benchmarks. Today we're testing it on a real app (Truth Lens) alongside Opus and GPT-5.4 to see if the benchmarks hold up.

We're also looking at Cursor's updated agent harness, their new interface (early access preview), and whether Composer 2 changes the daily workflow for developers already using Cursor with third-party models.


Join our Discord: https://rfer.me/discord


Key moments
Timestamps by https://raytranscribes.com
00:00:00 Cursor's New Composer 2 Model
00:01:15 TruthLens Hackathon App Introduction
00:03:03 Live Demo of TruthLens App
00:05:08 First Look at Cursor's Glass UI
00:08:05 TruthLens Architecture Deep Dive
00:15:19 Opus 4.6 vs. Composer 2 Setup
00:23:19 Comparing Opus vs. Composer Results
00:36:51 Real-Time Analysis of Podcast Clip
00:48:20 Inference Engineering Book
00:51:12 Rebuilding TruthLens with Composer 2
01:05:44 Testing the Composer-Generated UI
01:10:22 Discovering the Tavili Search API
01:18:52 Handbook of Neuroevolution Discovery
01:24:15 Final Thoughts on Composer 2
01:26:01 Cursor Glass UI Recap


CONNECT WITH RAY
X (Twitter): https://x.com/RayFernando1337


#CursorAI #Composer2 #AICoding #CursorComposer2 #AIModel #Claude #GPT #LiveCoding #Programming #SoftwareEngineering