β¨AI should be about empowering humans, building understanding, and making dreams realities. π©βπ» DevRel Eng. Lead @GoogleDeepMind ex-@GitHub || views = my own!
179Β°56β²39.4β³, +0Β°2β²46.2β³ 8k pc
Joined December 2008
- Tweets71.3K
- Following2K
- Followers76.7K
- Likes122K
Pinned Tweet
To get access to the latest Gemini models, including the most recent experimental release that's currently at the top of the @lmsysorg leaderboard, use @googleaistudio.
You can also turn on features like code execution; use extended context (2M+); and turn off safety settings: β at San Francisco, CA
β‘ Many folks have been using our Flash-Lite series and smaller Flash models as classifiers β excited to have this available as a tinier, speedier, and more cost-effective option!
Please try it out and let us know what you think: β at Georgetown, TX
Introducing the Gemini Decisions API
Powered by our new Gemini Decision 1 model, the API is built for fast routing, tool selection, safety checks, and grading inside your apps and agents without the latency or output-token cost of full text generation.
β’ Up to 1,000 questions evaluated per request
β’ $0.10/1M input tokens & $0 output tokens
β’ One shared input across text, images, audio, video, and PDFs
Start building today by grabbing an API key in @GoogleAIStudio
π Tiny, efficient, classification models still have utility! Love that we're making this available via the API, please test it out and let us know what you think. β at Georgetown, TX
Introducing Gemini Decision 1, our new frontier decision model in the Gemini API.
It is $0.10 per 1M input tokens with no additional cost. With support for image, audio, video, and PDF natively, 256K context window, & 1K questions.
You can try it for free in the API!
3
35
2,296
π©βπ» Paige Bailey retweeted
Insane jump from Argon to Carbon on Bug Bunt Bench using Antigravity CLI. Crazy how fast we are moving and how much more cost efficient the Gemini models are for the capability!
What a day! I tested Gemini 4 Carbon on real work. 2 repos, 105 replanted bugs. Find and fix what you can.
It scores just behind Fable 5.2. The results:
- Fable 5.2 (max): 58 for $139.65
- Gemini 4 Carbon (high): 57.3 for $31.32
- Sonnet 5.5 (max): 51.3 for $154.03
- GPT-6 Astra (max): 45 for $33.03
- GPT-6.1 Sol (max): 44.3 for $5.88
- Gemini 4 Argon (high): 40.7 for $7.16
Google's own benchmarks looked super strong, but nobody could access the model. Turns out they didn't lie: +16.6 points over Argon, at less than a quarter of Fable 5.2's cost.
Nobody should be laughing at Google anymore.
(the next effort level drops soon in the thread π§΅)
π©βπ» Paige Bailey retweeted
I've been digging through the #CoRL2026 program and am blown away by how robotics labs are wiring @GoogleDeepMind's foundation models into control loops, sim engines, and humanoids.
A thread of some of my favorite papers and posters from the event (+ where to find them!): π§΅π
Honestly, if a tiny robot dog physically walked up to my desk and stared at me until I drank water or stood up after a 4-hour coding session, my posture would be cured immediately. πππ§
Love seeing @GoogleGemma 4 running 100% locally on an @Apple iPhone (airplane mode!) to keep wearable health data completely private, all while driving @PetoiCamp's open-source Quaddle.
Incredible hackathon build! π
Your health companion has legs. π
Abhinav Agarwal, Rithika Gurram, and Shobhit Dhamania built Bolt at the AGI House(@agihouse_org) Gemma(@googlegemma) hackathon in Silicon Valley. A Clair Health bracelet streams live data to an iPhone, Google's Gemma 4 answers questions about it right on the phone, and Quaddle, our open-source robot dog, is the companion's body on the floor.
It works in airplane mode, with nothing sent to a server.
Bolt's code (MIT): github.com/abhinav30219/boltβ¦
Video by the Bolt team, who started from our Gemma + Petoi robots starter: starter repo: github.com/PetoiCamp/opencatβ¦
What would you like your AI health companion to do?
ccΒ @DynamicWebPaige @make @irvinxyz @WeeklyRobotics @VantixAgency
#PhysicalAI #Gemma #OnDeviceAI #Quaddle #OpenSourceRobotics #Robotics
π©βπ» Paige Bailey retweeted
Our recent Agent Arena analysis found that Gemini 4 Carbon by @GoogleDeepMind becomes more cost-competitive as tasks get more demanding.
Its savings relative to GPT-6 Astra and Claude Opus 5.5 grow on the most expensive sessions.
Models by @GoogleDeepMind scale better on cost than their rivals.
Gemini 4 Carbon is the prime example. As tasks get more demanding, every rival gets more expensive relative to Carbon.
For a typical task, GPT-6 Astra by @OpenAI costs 3.4Γ as much, Claude Opus 5.5 by @AnthropicAI costs 3.9Γ as much, and GPT-6.1 Sol is the cheaper option at about half of Carbonβs cost (0.54Γ).
Across the 5% most expensive sessions, Carbonβs position improves against all three. Opus edges up to 4.2Γ, Astraβs cost roughly doubles relative to Carbon, and Solβs discount shrinks.
This in turn can translate into meaningful savings across real-world tasks such as building an Android security app, one of many long-horizon tasks that our Agent Arena users delegate to their agents.
ππ Gemini 4 Carbon is officially here to solve for your Mohs hardest problems!
Puns aside, Carbon has been a gift - and is my daily driver for everything from knowledge work to webscraping + data analysis to fixing buggy software.
π Please test it out in @GoogleAIStudio, and let us know what you think!
Obsessed with both the name and the mission of OopsieVerse and OopsieData! β¨π€ #CoRL2026
Built on MuJoCo (via RoboCasa), DamageSim tracks object-centric "health" from physics signals like contact forces -- so we can train and benchmark robots not just on completing household tasks, but doing them without breaking stuff.
Awesome open-source work from @UTAustin:
π robin-lab.cs.utexas.edu/oopsβ¦
Decades of human effort β 1 month with AlphaProof Code and @GoogleDeepMind's Gemini Carbon. β‘
Imagine a near future where critical infrastructure and open-source libraries ship with mathematical proofs by default. Massive milestone from @pushmeet and team!
Announcing a major milestone achieved by my team @GoogleDeepMind: we have proven key security properties of the machine code underlying the Signal messaging protocol. This is the first time the actual production code of a widely deployed protocol has been formally verified β by any method, with or without AI β and at this depth.
More than a research breakthrough, itβs a prime example of how frontier AI can be a force multiplier to counter the speed and scale of modern cyber threats.
Formal verification offers a higher standard of security; however, itβs extremely complex and time consuming, which is why in many cases itβs considered to be impractical. Similar efforts could take teams decades of combined effort. We did it in a monthβs time.
Using AlphaProof Code, Google DeepMind's new agentic framework for formal verification that runs on LLMs including Gemini Carbon, we verified the Signal protocol down to the actual code running on devices.Β This milestone lays the groundwork for a future in which vital systems, from medical device firmware to power grid software, ship with mathematical proof that their protections work.
Reaching that future will take continued research from our community. In that spirit, we're open-sourcing our formal specifications and verification artifacts so others can build on this work.
On a personal note, this moment is of deep significance to me. I started my own research journey more than two decades ago as an intern working on checking the consistency of complex software systems. Seeing this grand challenge made practical with modern AI is extraordinary.
I'm extremely proud of my team members: Swarat Chaudhuri, Andrew Ferraiuolo, Anton Kovsharov, Tej Chajed, Sam Schlesinger,Β Karn Seth, Jonathan Katz, and Charles Sutton for taking on this grand challenge and delivering, and grateful to the Signal team for their partnership.
deepmind.google/blog/formal-β¦
π Come say hi! #CoRL2026 @GoogleDeepMind
π¦π₯Ή Tiny bat art in Zilker Park, celebrating all of Austin's native bats (and I quote!):
π "Mexican free-tailed bats β the stars of the show at Congress Avenue Bridge
π Cave myotis β beautiful bats with charm and charisma galore
π Evening bats β little bats with great big personalities
π Tricolored bats β the famed βChicken Nuggetsβ of Texas cave tours
π Eastern red bats β wild spirits of the woods
π Seminole bats β hang in loblolly pines, mimicking pinecones
π Northern yellow bats β large and lightning-fast palm tree dwellers
π Southern yellow bats β delicate, charming residents of palm trees
π Hoary bats β named for hoar frost, when fog freezes on windowpanes
Since bats' hands are their wings, when they fly, they are waving at us! Be Austin-friendly and wave back!"
@AustinBatRefuge
2
14
1,139
Honestly one of my favorite "devtool" hacks has been using @Antigravity for non-coding tasks -- analyzing datasets, digging through stacks of PDFs, helping answer emails, scraping user feedback, and general knowledge work.
Once you have an agent for everyday workflows, there's no going back. *Super, super excited* we're making it officially frictionless for everyone to work this way.
Huge congrats to @_mohansolo @rodydavis @kevinhou22 and team! β¨
