← Back to The Brief

05

Google folds computer use into Gemini 3.5 Flash as a built-in tool

incrementalDeveloperEnterprise

Thursday, June 25, 2026

Confidence

Low · — OSWorld score is self-reported and unreproduced

Evidence

DeepMind product blog + vendor-published benchmark

What happened

Google said computer use is now a built-in tool supported in Gemini 3.5 Flash, delivering its best performance yet for agentic computer use tasks; previously only available as a standalone Gemini 2.5 computer use model, computer use is now integrated natively in the main Gemini Flash model. Gemini 3.5 Flash achieves a 78.4 score on OSWorld-Verified tasks, matching premium models like Sonnet 4.6 while drastically outperforming Gemini 3 Flash at 65.1. See the DeepMind announcement and Neowin coverage.

What it means

Verdict held at INCREMENTAL because the capability already existed in a standalone model — what changed is packaging and price. That packaging matters: developers no longer pick between a Flash-class model and a Computer Use model, so the default agent on Google's API now has eyes and hands. The OSWorld parity with Anthropic's Sonnet 4.6 at Flash pricing is the competitive move; Anthropic's Computer Use premium just got harder to defend on capability alone.

Google said computer use is now a built-in tool supported in Gemini 3.5 Flash, delivering its best performance yet for agentic computer use tasks; previously only available as a standalone Gemini 2.5 computer use model, computer use is now integrated natively in the main Gemini Flash model. Gemini 3.5 Flash achieves a 78.4 score on OSWorld-Verified tasks, matching premium models like Sonnet 4.6 while drastically outperforming Gemini 3 Flash at 65.1. See the DeepMind announcement and Neowin coverage.

Sources

On this page

What happened
What it means

In this issue

← Back to brief

Stat of the Day

~50%▼

OpenAI inference cost, per Bloomberg

The cost reduction Jalapeño is reported to deliver versus current alternatives — the number that turns OpenAI's chip move from vanity into pricing leverage. VentureBeat

Today’s Take

Three of today's five stories — OpenAI's chip, Anthropic's letter, and Backbase's Kasisto buy — describe the same move from different angles: frontier labs and platform vendors reclaiming the layers their customers used to assemble themselves. The FDA designations point the other direction, opening a regulated path for new entrants in healthcare. The pattern that survives the week is this: where regulators do not write the rules, the platform owners do, and they are writing them now. The bet that is not working is "we'll stay portable across vendors" — vendor portability is being designed out of the stack on every side.

— Agentic desk

Never miss a signal

Join professionals getting the daily AI brief every weekday morning.

No spam · Unsubscribe anytime

No sponsored verdicts
Sources linked
Corrections public