Gemini 3.1 Pro preview achieves SOTA performance across benchmarks

Hey there!

Welcome back to The Pulse, where we dive into interesting AI stories and trends backed by data, all presented through simple visuals.

> ranks #1 across major benchmarks, including the Artificial Analysis Intelligence Index (aggregate of 10 benchmarks)

> significantly better performance & benchmark scores than latest Opus 4.6 (regarded as the “best”)

> now leads ARC-AGI 1 & 2: 3.1 Pro saturates ARC-AGI 1 at 98% & pushes the Pareto frontier on ARC-AGI 2 at $0.96 per task, at least 2× cheaper than other frontier models

> currently strongest overall for reasoning & coding, with improved factual grounding + reduced hallucinations

> however, Opus 4.6 still preferred widely for long agentic coding tasks, reflected in better performance in select agent benchmarks

> priced at $2/m input & $12/m output tokens - roughly half the cost of Claude & still below GPT-5.2

> thinking modes expanded from low/high to low/medium/high; on “high,” behavior reportedly resembles a lighter version of recently released Deep Think

> made available across the Gemini app, API, NotebookLM, etc.

> #1 in AA image arena at half the price of Nano Banana Pro; #3 in image editing

> #1 in LM text-to-image arena; ties #1 with ChatGPT in LM arena for image editing

> built on Gemini 3.1 Flash Image; combines Flash-speed efficiency with higher visual fidelity

> integrates Gemini + real-time web search info to make images

> features: advanced world knowledge, precise text rendering/translation, 512px→4K upscaling, aspect ratio control, subject consistency (up to 5 characters, 14 objects)

> replacing Nano Banana Pro in the Gemini app; available across Gemini, AI Studio, API, Search, etc.

> freelancers early casualties as firms shift pend from online labor platforms (0.66% → 0.14%, 2021–2025) to AI providers (0% → 3%)

> over half of freelancer-using firms in 2022 stopped by Q2 2025

> $1 human labor replaced by ~$0.03 of AI (20–25x cheaper); firms most reliant on freelancers adopted AI first

> adoption is widespread: 93% of developers use AI assistants monthly & 26.9% of production code is AI-generated (early 2026)

> yet productivity gains stall near ~10%, 80% of 6K executives see no impact; GS estimates AI added “basically zero” to U.S. GDP in 2025

> likely due to individual adoption instead of org-level integration