Gemini 3.5 Flash gets computer use, and its own chart undercuts it
Google folded screen-driving agents into its cheap Flash model. Commenters opened Google's own OSWorld graph and found Flash trailing Opus 4.8 and GPT 5.5.
Computer use is now native in Gemini 3.5 Flash: the model sees a screenshot, reasons, and acts across browser, mobile and desktop, pitched for software testing and enterprise automation via the API and a Browserbase demo.
Then the thread turns on the marketing. Readers note the headline benchmark chart visually crowns Flash while the numbers show two rivals ahead, and agent builders argue screenshot-driving is the wrong primitive entirely.
Screen-driving agents, now built into the cheap model
Google's announcement makes computer use a native, built-in tool in Gemini 3.5 Flash rather than a separate standalone model: the agent can "see, reason and take action across browser, mobile and desktop environments." The pitch is long-horizon work, continuous software testing and knowledge tasks across professional apps, reachable through the Gemini API, the Enterprise Agent Platform and a Browserbase-hosted demo. Google leans on OSWorld as the headline benchmark and frames safety around adversarial training plus optional enterprise guardrails for sensitive actions.
The benchmark chart becomes the story
The first thing the thread did was open Google's own graph and read it against the framing.
"Gemini 3.5 Flash is beat hands down by both Opus 4.8 and GPT 5.5, and yet the graph is drawn as if Gemini wins." — a commenter reading Google's own OSWorld chart
The sharpest technical objection came from someone shipping a competing agent who thinks the whole screenshot paradigm is wrong.
"we construct custom accessibility trees… switched over to using DeepSeek v4 Flash and its nearing 100x cost decrease… screenshots to take actions on a webpage… seems too naive." — the founder of web-agent startup rtrvr.ai
A broader doubt questioned whether computer use is a good idea at all.
"Computer use is such a terrible idea. It's slow, insecure, error prone, expensive." — a commenter on the ergonomics of LLM-OS interfaces
And the trust problem showed up as a concrete near-miss when someone let Flash touch a repo.
"Gemini Flash 3.5… ran
git reset --hardwhen I asked it to commit my changes… Of course I'm not trusting my computer to it." — a user who handed it a git task
The defense was narrow but real: at roughly a third the cost of GPT 5.5, Flash is the model you reach for when speed and price beat the leaderboard. The launch's actual tension isn't capability. It's that Google shipped an agent to drive your desktop using the model people trust least to be right.
Get breakdowns like this every day
The top Hacker News stories, unpacked — by email, Telegram, or RSS.
An administration letter to Anthropic locked the NSA out of Mythos too
A directive meant to restrict a powerful code-auditing model swept so wide that even Anthropic's own staff lost access. The spy agency that wanted it most was caught in the same net.