ByteDance Seed: 2.0 Code vs 2.1 Turbo
Two new ByteDance Seed models went through the matrix together — both 35 of 35 on build reliability, but with opposite temperaments.
Seed 2.1 Turbo is the metronome: a 3.26 average with only four builds below three stars. Matrix rain, the Mandelbrot explorer, the Lorenz attractor, boids, and a full-screen Game of Life all land clean. The code-tuned Seed 2.0 swings harder at 2.71 — it matches Turbo's four-star ceiling (13 builds each) but nine outright-broken outputs drag the average down: a blank double pendulum (`cos1 is not defined`), a force graph that errors before it initializes, an aquarium that never fills.
The interesting part is that they break on different prompts. Turbo fumbles the brick-breaker and the maze that Seed 2.0 gets right, while Seed 2.0 blanks on the pendulum Turbo animates cleanly. For these visual one-shots, the "code" model is the streakier one — higher variance, not higher output. The paired builds are below.
