Gemini 4 Carbon vibe check: boldest UI designs, least reliable finishes
every · x · 2026-11-17
Every's vibe check of Google DeepMind's Gemini 4 Carbon, which launched a day after Anthropic's Fable 5.2 raised the flagship bar.
- In blind tests, tester tylernishida found Carbon produced the most varied UI patterns of any model, including design approaches he hadn't seen elsewhere — but it failed to finish several builds, sometimes needing up to 3 "continue" prompts.
- Versus Gemini 4 Argon it's a much stronger coder, makes better 3D objects, and can run ambitious builds for hours with little supervision.
- The key testing framework: evaluate design direction and reliable delivery independently. Carbon codes better, but Fable showed an agent should also pick the right work, respect assignment limits, and know when it's done.
- The full review is subscriber-only; Google provided early access with no editorial input.
More from coding & agent
- 39 open-weight models, 11 days, 8,199 runs: 75.7% of agent failures came from runs that did call tools — Green_Competition_21 · 2026-11-17
- gputaskman: an open-source htop for your GPU that breaks down VRAM by process and LLM KV-cache — alex_verem · 2026-11-17
- Matt Pocock's /constrain Idea: Cage Agents With Brutal Lint Rules — mattpocockuk · 2026-11-17
- Editing WordPress via MCP Feels Like "Superpowers"; Contributors Want It in Core — VeryWellVersed · 2026-11-17
- Self-hosted MCP endpoint exposes 38 database engines, read-only, creds stay server-side — Green_Competition_21 · 2026-11-17
- Stop Over-Trusting AI Coding Leaderboards: CACM Paper Cites OpenAI's SWE-bench Verified Reversal — EhudReiter · 2026-11-17