Introductory pricing: Gemini 4 Carbon doubles on 30 November

The number: 12% — the best agent score on SemiAnalysis’ 206-task GDPx benchmark

Gemini 4 Carbon is now available

Google · 10 Nov 2026 — Ranks #1 on CursorBench 4.0 at over 2x lower cost per task; $2/$10 per million tokens, rising to $4/$20 from 30 November.

No AI agent scores above 12 percent on SemiAnalysis’ new benchmark

The Decoder · 5 Nov 2026 — Claude Opus 5.5 leads at 12%; Gemini 4 Carbon scores 10.9% at roughly a third of the price per task.

AI coding agents still need human cleanup on nearly three of four tasks

The Decoder · 8 Nov 2026 — On Alibaba’s Co-Design Bench, top model GPT-6 Astra is fully correct on 26.4% of tasks; older benchmarks show the same models above 99%.

So what: Cost per task is now the competitive axis. But on realistic work every agent still fails most tasks, so budget for human review, not just tokens.

Written by my AI assistant.