Every's vibe check: Gemini 4 Carbon scores 81 at low thinking vs 54 at high under deadlines
every · x · 2026-11-11
Every tested Gemini 4 Carbon on 9 real-world coding tasks with a 10-minute limit: low thinking averaged 81/100, medium 62, high 54, with 5 runs hitting the time limit and 3 scoring zero. Without deadlines the gap nearly vanished (81/76/80), suggesting higher thinking just costs time under real schedules. The review finds Carbon a far stronger coder than Gemini 4 Argon that can build for hours unsupervised, but Anthropic's Fable 5.2 has raised the bar for agent judgment and knowing when a task is done.
Related event: Gemini 4 Carbon Launches: Early Tests Show Big Coding Gains Over Argon(10 posts)→
More from coding & agent
- Claude agent drives a real Leica microscope via MCP, autofocus and all — jwt0625 · 2026-11-11
- Yacine's agent talk: context moves outside the model, harnesses self-edit — yacinelearning · 2026-11-11
- UluP Nexus: a read-only MCP server for discovering indie software with verified submissions — Potential-Art7696 · 2026-11-11
- Cursor ships multiplayer in Projects: shared task board with agent-to-agent requests — mattyp · 2026-11-11
- Why agents will soon have only one tool: Code Mode cuts spend 15%, +3.5 Terminal-Bench points — dl_weekly · 2026-11-11
- MCP Stock Assistant serves real-time A-share and US stock data via Xueqiu API — modelcontextprotocol · 2026-11-11