Gemini 3.1 Pro: Google’s Abstract Logic and Multimodal Reasoning Architecture Resets the Scientific AI Benchmark (March 2026)
Google's Gemini 3.1 Pro achieves 94.3% on GPQA Diamond and 77.1% on ARC-AGI-2 at $2/12 per million tokens — but NotebookLM regressions expose benchmark optimization trade-offs.
Read story