グーグル「Gemini 3.1 Pro」リリース 推論性能が「3 Pro」の2倍以上 – ケータイ Watch https://www. yayafa.com/2781307/ # A
agents deepmind gemini google
| Source: Mastodon | Original article
Google has rolled out Gemini 3.1 Pro, the latest upgrade to its Gemini family of large‑language models. According to the company, the new model delivers more than twice the inference speed of its predecessor, Gemini 3 Pro, and pushes its ARC‑AGI‑2 benchmark score to 77.1 percent – a level that now exceeds the average human score on the same test. The leap in raw reasoning ability is paired with a 1 million‑token context window, enabling the model to handle sprawling codebases, extensive data tables and multi‑page documents without truncation.
The performance boost matters because it narrows the gap between specialised AI systems and a truly “agentic” assistant capable of tackling end‑to‑end workflows. Early demos show Gemini 3.1 Pro generating SVG animations on the fly, assembling real‑time analytics dashboards, and writing production‑grade code in languages ranging from Python to Rust. Its multilingual competence extends to advanced mathematics and programming tasks in Japanese, Korean and other non‑English languages, a claim that sets it apart from many Western‑centric competitors.
Pricing is another strategic lever: Google’s API rates for Gemini 3.1 Pro are positioned at roughly half the cost of OpenAI’s GPT‑4o, while benchmark comparisons published by independent analysts place the model ahead of GPT‑5.3 and Anthropic’s Claude Opus 4.6 on complex reasoning and multi‑modal tasks. If the figures hold up in broader deployments, enterprises could adopt Gemini 3.1 Pro for heavy‑duty AI workloads that previously required a patchwork of specialised tools.
What to watch next are the integration pipelines Google will open for developers. The company hinted at tighter coupling with its Cloud Vertex AI platform and upcoming support for “agentic” tool use, echoing the broader industry shift toward AI that can invoke external APIs, retrieve live data and execute actions autonomously. The next few months will reveal whether Gemini 3.1 Pro can translate its laboratory scores into sustained productivity gains for developers, data scientists and business users across the Nordics and beyond.
Sources
Back to AIPULSEN