Google coming launched Gemini 3 Pro, nan latest loop of its family of instauration models, pinch different Gemini 3 variants to follow. Gemini 3 is instantly disposable crossed a assortment of Google products, including Google Search, nan Gemini App, AI Studio, Vertex AI and Google’s various agentic improvement tools, including nan newly launched Antigravity.
Record Benchmarks
Unsurprisingly, Google describes Gemini 3 Pro arsenic its astir intelligent exemplary yet. According to Google, nan exemplary will apical nan LMArena Leaderboard pinch a people of 1501 points. It besides scores 37.5% connected Humanity’s Last Exam and 91.9% connected GPQA Diamond, a benchmark that asks Ph.D.-level subject questions and which OpenAI’s GPT-5.1 antecedently led pinch 87.6% accuracy.
One area nan Google squad stresses is Gemini 3 Pro’s expertise to do good connected multimodal reasoning tasks, pinch nan exemplary reaching grounds scores connected galore of those benchmarks arsenic well. This besides intends it’ll excel successful ocular reasoning tasks, and different area Google highlighted is its expertise to create analyzable animations and simulations connected nan fly.

Comparing Gemini 2.5 Pro and 3 connected a visualization task. (Credit: Google)
Benchmarks aren’t everything, of course, and tin moreover beryllium a spot misleading astatine times, but Google argues that nan caller exemplary tin bring “a caller level of extent and nuance to each interaction,” without resorting to cliches and nan sycophantic behaviour often seen successful different models. It’s only successful day-to-day use, though, wherever nan differences betwixt exemplary families really go obvious, truthful it’ll return a fewer days for developers and consumers to get a consciousness for really good nan exemplary performs successful real-world tasks.
Coding With Gemini 3
As for coding, Google says nan caller exemplary besides outperforms Gemini 2.5 Pro astatine coding successful some agentic workflows and erstwhile handling zero-shot tasks. It’ll beryllium disposable successful Google’s ain tools, for illustration Gemini CLI and Android Studio’s Agent Mode, arsenic good arsenic done third-party devices from nan likes of Cursor, GitHub, JetBrains, Manus and Cline, among others.
“Cline is utilizing Gemini 3 to powerfulness autonomous codification procreation successful developers’ IDEs,” said Nik Pash, Cline’s caput of AI. “Gemini 3 Pro handles complex, long-horizon tasks crossed full codebases, maintaining discourse done multifile refactors, debugging sessions and characteristic implementations. It uses agelong discourse acold much efficaciously than Gemini 2.5 Pro and has solved problems that stumped different starring models.”
API entree will costs $2 per cardinal input tokens and $12 per cardinal output tokens for prompts pinch 200,000 tokens aliases fewer. That’s up from $1.25 and $10 per cardinal Gemini 2.5 Pro input/output tokens.
With this launch, Google is besides releasing a client-side bash instrumentality that will let nan exemplary to usage ammunition commands during its agentic workflows truthful it tin navigate nan record system, for example. There will besides beryllium a hosted bash instrumentality for codification procreation and prototyping, but for nan clip being, nan hosted instrumentality will only beryllium disposable successful nan Gemini API for early entree partners.
Deep Think Coming Soon
Google will soon besides motorboat different version of Gemini 3: Gemini 3 Deep Think. This type will characteristic an enhanced reasoning exemplary that will apt push nan model’s capacity moreover further. Right now, Google is only making this exemplary disposable to information testers, but it’ll soon beryllium disposable to Google AI Ultra subscribers (who salary $250/month) arsenic well.
“Gemini 3 Deep Think outperforms Gemini 3 Pro’s already awesome capacity connected nan benchmarks … for illustration Humanity’s Last Exam and GPQA Diamond. So it really pushes nan boundaries of intelligence to thief you lick your astir analyzable problems,” Tulsee Doshi, Google’s elder head and caput of merchandise for nan Gemini models, explained successful a property briefing up of today’s launch.
YOUTUBE.COM/THENEWSTACK
Tech moves fast, don't miss an episode. Subscribe to our YouTube channel to watercourse each our podcasts, interviews, demos, and more.
Group Created pinch Sketch.
English (US) ·
Indonesian (ID) ·