Google says Gemini 3.5 Flash rivals 'large flagship models' for coding and agentic tasks

Trending 2 weeks ago

It tin complete tasks successful a "fraction of nan time" of different frontier models, Google claims.

A Google floor plan showing Gemini 3.5 flash vs Gemini 3.1 pro for coding, agentic and scaled instrumentality use

Google

Google has unveiled Gemini 3.5, starting pinch nan Gemini 3.5 Flash exemplary that promises to outperform Gemini 3.1 Pro successful real-world agentic and coding tasks. "3.5 Flash delivers frontier-level intelligence astatine exceptional velocity — proving you nary longer person to waste and acquisition value for latency." 

Announced at Google I/O 2026, this will beryllium Google's default AI exemplary (not to beryllium confused pinch Flash-Lite), designed to present amended velocity than nan existent Gemini Pro models astatine a much affordable price. The tradeoff is little capacity than nan 3.5 Pro exemplary (coming adjacent month) successful tasks that require heavy reasoning and high-context understanding. 

However, Google has reduced nan discuss betwixt nan Pro and Flash models, saying Gemini 3.5 Flash "delivers intelligence that rivals ample flagship models connected aggregate dimensions." It outperforms nan existent Gemini 3.1 Pro exemplary connected coding and agentic benchmarks for illustration Terminal-Bench 2.1 (76.2 percent), MCP Atlas scaled instrumentality usage (83.6 percent) and multimodal knowing pinch 84.2 percent connected CharXiv Reasoning. From an output tokens-per-second standpoint, it's 4 times faster than different frontier models, according to Google. 

Google floor plan showing Gemini 3.5 flash output velocity compared to Gemini 3.1 ProGoogle

All of that makes 3.5 Flash perfect for long-horizon agentic tasks, completing what utilized to return weeks successful "a fraction of nan time," Google wrote. "Under supervision, it tin reliably execute multi-step workflows and coding tasks while sustaining frontier performance." It added that partners, including banks and fintechs, person utilized it to automate multi-week workflows.

Google noted that 3.5 Flash is now nan default exemplary for nan Gemini app and for AI Mode successful Search worldwide. The individual AI supplier Gemini Spark, rolling retired to testers today, besides runs connected 3.5 Flash.  At nan aforesaid time, Google said it strengthened Gemini 3.5's cyber and CBRN (chemical, biological, radiological, nuclear) safeguard truthful it's little apt to make harmful contented (or mistakenly garbage to reply safe queries). Google elaborate nan caller model's enhanced agentic coding capabilities successful a new blog post.

Gemini 3.5 Flash is now "generally disposable via Google Antigravity, nan Gemini API successful Google AI Studio and Android Studio, Gemini Enterprise Agent Platform and Gemini Enterprise. It's besides now disposable to everyone successful nan Gemini app and AI Mode successful Search," Google said. 

There's nary connection yet connected pricing for paid tiers. As for Gemini 3.5 Pro, Google said: "It's already being utilized internally, and we look guardant to rolling it retired adjacent month."

More
Source engadget.com
engadget.com