Hacker News AI · 2026/10/14 00:03:16
Google 发布 Gemini 4 Argon：主打前沿科学难题与多领域专家协作能力
Google 正式推出新一代旗舰模型 Gemini 4 Argon，定位为“解决最棘手问题的前沿智能”。该版本重点强化了科研、工程、安全等领域的专家级协作能力，旨在突破科学与数学边界。目前已在 Gemini、AI Studio 及 Antigravity 平台开放体验。
报道全文原始报道全文
本文目录23 个章节
Try in GeminiTry in Google AI StudioTry in Google Antigravity
Gemini 4 Argon
Our next era of frontier intelligence, built to solve your hardest problems
Your browser does not support the video tag. Your browser does not support the video tag.
Explore the latest
Slide 1 of 4
Solving frontier problems with experts
Researchers, engineers, and security specialists are pairing with Argon to push the boundaries of science, math, robotics and cyber defense.
Unlocking new scientific breakthroughs
Argon is helping researchers at Google and the scientific community accelerate their path to discovery and real-world impact.
Accelerating robotics
Argon autonomously turns raw video to a physics-accurate simulation, estimating depth and physical boundaries, and writing the exact simulation code required to replicate those real-world environments.
Scaling cybersecurity defense responsibly
Argon ties for #1 on CWE-Bench (68%) for vulnerability remediation, is our most resilient model against indirect prompt injections, and is expanding to hundreds of additional Fairwind Program partners today.
Solving complex engineering tasks
Across Google and our early access partners, Argon is quickly becoming the preferred model for engineers tackling long-horizon software engineering.
Your browser does not support the video tag.Untangling browser feature code in ChromeQuantum chemistry open-source package optimizationOptimizing a production Python codebase in hoursMaking our open Gemma models nearly three times as fastMigrating legacy code
Argon helped the Chrome team separate core webpage handling from a ~6 million line legacy folder of browser features held within Chromium’s 40 million-line codebase. Piping multiple clean code changes at a time, Argon helped the team finish an architectural refactor across hundreds of files, completing months of work in weeks. For Chrome's billions of global users, this will unlock the ability for live webpages to move fluidly between tabs, side panels, and floating windows without reloading, eliminating an entire category of browser crashes.
xTB (Semiempirical Extended Tight-Binding) is one of the most-cited open-source quantum chemistry packages for molecules. Scientists use it to predict molecules' 3D shapes and to screen large numbers of drug and materials candidates. Using Argon in Antigravity, our researchers made xTB up to 2.3x faster at predicting molecular structures and 1.8x faster at conformer search. The changes have been contributed to the official xTB repository on GitHub.
Argon in Antigravity is helping the team behind Google's open-source Agent Development Kit (ADK), a framework powering enterprise agents, push their performance engineering further. The team gave Argon their .agents/skills rulebook and a one-paragraph goal to eliminate framework overhead without breaking existing apps. Working autonomously overnight, Argon surfaced optimizations, now live in google/adk-python, that cut in-memory session prep time by 40% in long sessions and sped up cold imports by 21%.
Using Argon, we achieved a 2.85x end-to-end inference speedup on an A100 via vLLM (accelerating throughput from ~37,900 to over 108,000 tokens per second on Gemma 4 E2B) with 100% token parity and zero quality drop in just a few days, a process that would usually take months. Check out the upstream pull requests (optimizing everything from attention tuning to speculative decoding) to see the exact changes.
Argon in Antigravity migrated a backlog of 700+ Google Skills training labs off of hardcoded system images and runtimes spread across five code repositories, catching 16+ hidden timeout and race-condition bugs on the way. The work saved 955 engineering hours and reduced recurring cleanup sweeps from roughly 540 hours a year to a single 3-hour update.
Delivering complex 3D visualizations
Argon brings major improvements in spatial reasoning, mathematical precision, and visual coding compared to previous Gemini models.
Slide 1 of 5
Your browser does not support the video tag.
Argon built a highly detailed model of a Pagoda using multi-agent visual critique loops in Antigravity. The model autonomously sculpted a multi-building temple complex with photorealistic architecture, foliage and lighting in Blender.
Your browser does not support the video tag.
Argon turned pictures of a real desk into a playable miniature racing game using a multi-agent loop in Antigravity. The desk objects, room and race track were modeled and collision-checked using Blender MCP, textured with Nano Banana, and wired with physics, custom shaders, and car engine audio from Gemini Omni Flash in Three.js.
Your browser does not support the video tag.
Argon in Antigravity wrote Python scripts to procedurally sculpt, rig, animate, and PBR-bake a complex 3D game using multi-agent visual critique loops then wired it for the web in Three.js with GLSL shaders, shatter physics, and a single-player AI.
Your browser does not support the video tag.
With zero external libraries and 100% pure Python, Argon wrote its own 3D ray-tracing engine with a custom video tool and audio mixer to build and document a sunlit library from scratch, producing, narrating, and assembling this video with Gemini 3.8 TTS voiceover and music by Lyria 3.5.
Your browser does not support the video tag.
Starting from a single Nano Banana image in Antigravity, Argon drives Blender MCP in an autonomous loop, aligning camera grids, sculpting Gothic architecture, and staging audio-reactive lighting, to create a finished 3D scene.
Your browser does not support the video tag.
Argon built a highly detailed model of a Pagoda using multi-agent visual critique loops in Antigravity. The model autonomously sculpted a multi-building temple complex with photorealistic architecture, foliage and lighting in Blender.
Your browser does not support the video tag.
Argon turned pictures of a real desk into a playable miniature racing game using a multi-agent loop in Antigravity. The desk objects, room and race track were modeled and collision-checked using Blender MCP, textured with Nano Banana, and wired with physics, custom shaders, and car engine audio from Gemini Omni Flash in Three.js.
Your browser does not support the video tag.
Argon in Antigravity wrote Python scripts to procedurally sculpt, rig, animate, and PBR-bake a complex 3D game using multi-agent visual critique loops then wired it for the web in Three.js with GLSL shaders, shatter physics, and a single-player AI.
Your browser does not support the video tag.
With zero external libraries and 100% pure Python, Argon wrote its own 3D ray-tracing engine with a custom video tool and audio mixer to build and document a sunlit library from scratch, producing, narrating, and assembling this video with Gemini 3.8 TTS voiceover and music by Lyria 3.5.
Your browser does not support the video tag.
Starting from a single Nano Banana image in Antigravity, Argon drives Blender MCP in an autonomous loop, aligning camera grids, sculpting Gothic architecture, and staging audio-reactive lighting, to create a finished 3D scene.
Transforming critical business and professional workflows
Argon combines frontier reasoning, deep visual perception, expanded domain knowledge, and stronger writing capabilities across legal, finance, science, and coding to handle high-stakes professional workflows.
Your browser does not support the video tag.
At Isomorphic Labs, the team is using Argon as a partner to accelerate their drug discovery and design research workflows. Argon is able to quickly synthesize complex PDFs and patents to extract pertinent information, reducing a process that typically takes weeks into hours.
Ambience HealthcareAmplitudeAxiom BioBoxCartwheelDeloitteKeplerLexisNexisMinutalAQuadrillonQuoraRoboflow
“On over 200 complex synthetic longitudinal charts, Gemini 4 achieved the highest recall and ICD-10 code specificity among the models we evaluated, producing concise problem lists without external lookups. That combination of medical knowledge and long-context understanding shows promise for clinician-reviewed chart workflows. We're now validating it on real patient records and excited to try it out more broadly.”
Brendan Fortuner
SVP Technology, Ambience Healthcare
“Argon is doing a great job at understanding what people actually mean. Customers collect endless signals that they don't know what to do with. Argon consistently scored amongst the top in our evals for building charts that help visualize the solutions for customers. It's also meaningfully persistent about finding those kinds of needle in a haystack answers that are valuable to customers.”
Curtis Liu
CTO, Amplitude
“Gemini 4 represents a step change in what AI can accomplish in real-world science. We've seen it reason through complex experimental data, from cellular imaging to mass spectrometry, and make subtle findings that traditionally require expert scientists to identify. On our Scientific Agent Benchmark, Gemini 4 was at the frontier across most tasks, including our hardest: ranking the clinical risk between two drugs.”
Alex Cabrera
Founding Engineer, Axiom Bio
“Gemini 4 Argon is at the frontier of enterprise knowledge work, achieving a 14% improvement in accuracy compared to Gemini 3.1 Pro. In Box’s evaluation, Argon connected evidence across the full set of documents, including details embedded in figures, and carried multi-step analyses through to complete answers. This is especially valuable in data analysis and in technical fields like healthcare and life sciences, where teams need to find and interpret information across many kinds of source material.”
Yashodha Bhavnani
VP of AI Products at Box
“By integrating Gemini 4 into our 3D spatial reconstruction engine, we turn everyday video into articulated, physics-ready 3D scenes in seconds instead of hours. Across our benchmarks, Gemini outperformed every model we tested — cutting runtime nearly 100× and token usage by 90%, unlocking physics ready and watertight 3D asset creation for our customers.”
Andrew Carr
Co-founder and Chief Scientist, Cartwheel
“Early impressions from Gemini 4 are very strong: for long-running tasks across these workflows, it completed successfully 100% of the time in our evals, maintaining context and running through multiple tools across impressive spans. Its depth of reasoning delivers significantly higher, clearer output. Performing in-depth reasoning for domain-specific tasks without detailed prior logic or instructions is something we’ve never seen with any model in its class before.”
Gopal Srinivasan
Global AI and Data Lead, Deloitte
“On our core financial analysis benchmarks, which span building multi-year financial statements straight from SEC filings to valuing companies with complex capital structures and announced acquisitions, Gemini 4 matched our analyst-built reference models on more than 99% of figures. Performance on our most complex workflows was at the frontier in a majority of tasks. We're excited to keep building with Gemini 4 at Kepler.”
Glenn LeBlanc
Software Engineer, Kepler
“To help legal professionals achieve more with Lexis+ with Protégé, we continuously evaluate AI models as part of the multi-model approach behind the LexisNexis Legal Intelligence Engine. In preliminary evaluations, Gemini 4 Argon demonstrated strong performance across legal knowledge, drafting, and analysis tasks, including identifying relevant citations and improving contract reviews, complaint analyses, and research memos.”
Min Chen
Chief AI Officer, LexisNexis Legal & Professional
“In early testing, Gemini 4 Argon stood out by connecting dates, documents, and financial calculations with legal precedents to develop highly substantive arguments. In one evaluation, it actually surpassed our human expert's answer key by surfacing a legally relevant detail we had missed. This demonstrates AI’s potential to help professionals uncover crucial details and bring a more complete understanding to complex cases.”
Caio Perona
CEO, MinutalA
“In Qualia, our autonomous AI researcher, Gemini 4 Argon has been a standout in our evaluations for both capability and efficiency. On our internal benchmarks of complex, multi-step research tasks, it performs at the frontier while completing our evals 1.5x faster than peer models at a quarter of the cost per task. It excelled at coming up with novel ideas and identifying non-obvious insights about the problem, helping it maintain a clear research vision over long horizons, something today's models struggle with.”
Trenton Chang
Chief Scientist, Quadrillion
“Gemini 4 impressed us with the quality of its responses in our early evaluations. On Poe, people are increasingly using AI to build things, and Gemini 4 gives creators a powerful foundation for turning an idea into a tool they can refine and share.”
Gareth Jones
Head of Product, Quora
“Argon ranked first in our object detection evaluations and led our new text benchmark, which uses real-world images from manufacturing, logistics, and aerial imagery. We’ve fully integrated Argon into Roboflow so developers and enterprises can leverage its high precision detection and text extraction to inspect products, track inventory, and guide machines.”
Brad Dwyer
Co-founder and CTO of Roboflow
Slide 1 of 4
Your browser does not support the video tag.
In Gemini App, Argon takes an unstructured folder full of project context and formats it into a clean proposal deck for a client using a custom skill. Gemini then drafts and emails the deck to a client while creating an automated workflow for all future incoming client requests with a scheduled task.
Your browser does not support the video tag.
See how Argon synthesizes ~50 GB of Airborne LiDAR scans, GIS room vectors, Library of Congress CAD drawings, and NPS archives to reconstruct 3D models of Chaco Canyon, turning fragmented archaeological records into explorable digital heritage sites.
Your browser does not support the video tag.
Argon using Antigravity in the Gemini macOS app generates design iterations for new legs for a sofa using Nano Banana then generates a 3d printable file for one. Starting from a sketch of the measurements and a photo of the couch Argon builds the model to the correct size to fit the sofa.
Your browser does not support the video tag.
Replace a broken sofa leg with new 3D printed ones using custom visualizations using Antigravity in the Gemini macOS app.
Your browser does not support the video tag.
In Gemini App, Argon takes an unstructured folder full of project context and formats it into a clean proposal deck for a client using a custom skill. Gemini then drafts and emails the deck to a client while creating an automated workflow for all future incoming client requests with a scheduled task.
Your browser does not support the video tag.
See how Argon synthesizes ~50 GB of Airborne LiDAR scans, GIS room vectors, Library of Congress CAD drawings, and NPS archives to reconstruct 3D models of Chaco Canyon, turning fragmented archaeological records into explorable digital heritage sites.
Your browser does not support the video tag.
Argon using Antigravity in the Gemini macOS app generates design iterations for new legs for a sofa using Nano Banana then generates a 3d printable file for one. Starting from a sketch of the measurements and a photo of the couch Argon builds the model to the correct size to fit the sofa.
Your browser does not support the video tag.
Replace a broken sofa leg with new 3D printed ones using custom visualizations using Antigravity in the Gemini macOS app.
Accelerating real-world impact with researchers and domain experts
Argon is built to be an active research partner, moving beyond literature search to help scientists, mathematicians and researchers formulate, stress-test, and prove non-obvious hypotheses.
Quantum computingNuclear fusionRare disease diagnosisEnzyme design
Argon is helping our researchers accelerate progress across both quantum error correction and algorithm design. To help error decoders keep pace with quantum hardware in real-time, Argon translated an internal version of AlphaQubit, our ML quantum error correction decoder, into a single, highly optimized hardware kernel – nearly doubling its inference throughput so we can deploy larger, more capable models and scale to larger quantum devices, while continuing to meet strict real-time thresholds. Argon is also reducing spacetime costs for quantum subroutines. For arcsin, a building block in recent quantum simulation algorithms, an implementation written by Argon and formally verified in Lean cuts spacetime cost by more than 100x compared to a published quantum CORDIC baseline, a step toward a standard library of optimized quantum circuits.
Stellarators are incredibly complex machines that represent a leading candidate for limitless clean fusion energy, but optimizing their 3D magnetic fields is famously difficult. Since 1967, a long-standing hypothesis known as “Grad’s conjecture” assumed that finding exact mathematical solutions for these fields was impossible, until recent counterexamples were found, including by Professor Matt Landreman (University of Maryland). Now, a team at Google alongside Professor Landreman has discovered a broader class of counterexamples to Grad's conjecture. These new solutions give the fusion community reliable benchmarks to test their stellarator simulations codes, which are valuable tools for future designs.
Researchers at Gamow Labs used Argon to analyze de-identified genomes from newborns with rare developmental lung disorders. In one case, Argon pinpointed a chromosomal rearrangement that routine software missed for years, and explained why it went undetected. On private Gamow Labs evals derived from real clinical and lab data, Argon topped on RareBench (57% on assistance with rare-disease diagnosis) and LabBench (52.2% on wet-lab reasoning). These advances point to faster diagnosis for families facing rare conditions.
Engineering enzymes with precise reactivity is notoriously slow, often demanding months of bespoke model tuning and candidate screening. Serine esterases are a class of enzymes that cut ester bonds on targeted molecules, widely used in recycling, and the manufacture of pharmaceuticals, biofuels and detergents. Our scientists worked with Argon to design serine esterases, targeting the benchmark molecule 4MU-acetate that glows under UV light once its ester bonds are broken. In a matter of hours, Argon assembled relevant data, performed model tuning and identified multiple biochemistry-inspired filters to get to a final set of five novel designs. Subsequent laboratory testing confirmed robust catalytic activity across these designs. The speed and ease of successful enzyme generation demonstrates the potential for rapid acceleration in enzyme engineering in fields such as biomanufacturing, green chemistry, and drug development.
Speeding up robotics development
By bringing together advanced visual-semantic reasoning, deep physical understanding, and autonomous code generation, the model introduces a new set of capabilities across the robotics stack.
Your browser does not support the video tag.
To demonstrate these capabilities in action, we tested Argon’s ability to learn an out of distribution physical skill simply by watching. In an internal test, Argon watched a raw video of a human playing cornhole, mapped the spatial coordinates of the environment, and calculated the required whole-body joint movements. It then autonomously built the simulation code. When transferred onto Apptronik’s Apollo humanoid, the robot quickly sank seven consecutive shots in a real-world trial.
Gemini 4 Argon with cybersecurity defense capabilities
Bringing state-of-the-art cyber capabilities to more trusted defenders
Your browser does not support the video tag.
Wiz is already using Argon for its “Scan for Good” initiative, which protects critical public infrastructure for free. Having had early access to the model, Wiz used Argon to protect critical healthcare infrastructure. The model went beyond just spotting a vulnerability - it proactively safeguarded hundreds of thousands of patient records.
Unlocking new applications with a 1M-token output window
Uninterrupted frontier math and science derivations
With 1M output tokens, Argon has the headroom to finish 100K+ token reasoning trajectories and output exhaustive, step-by-step mathematical proofs, quantum error-correction decoders, and complete computational physics solvers in one go.
Full-repository generation and refactoring
Generate entire multi-file applications, scaffold complete microservices, or refactor sprawling codebases in a single pass without stitching together truncated responses.
Massive legacy translations
Translate monolithic legacy files and complex database schemas end-to-end while preserving cross-file dependencies.
Exhaustive structured synthesis
Produce comprehensive compliance audits, full-scale technical specifications, and massive structured datasets in a single turn.
Performance
Frontier performance in complex workflows
| Category | Benchmark | Notes | Gemini 4 Argon | GPT-6 Astra | Claude Fable 5.1 | Claude Opus 5.5 |
|---|---|---|---|---|---|---|
| Knowledge work | Vals Index | 68.9% | 63.1% | 65.8% | 67.0% | |
| AutomationBench | Score | 51.3% | 41.4% | 31.4% | 42.5% | |
| Vals Finance Agent v2 | 65.4% | 53.5% | 58.9% | 58.6% | ||
| Harvey's Legal Agent Benchmark | 19.6% | 5.4% | 6.7% | 3.8% | ||
| Agentic coding | DeepSWE v1.1 | 77.9% | 74.1% | 67.4% | 74.2% | |
| FrontierSWE v2 | 55.0% | 65.5% | 56.3% | 62.3% | ||
| Vibe Code Bench | 91.9% | 89.6% | 90.3% | 90.3% | ||
| Terminal-bench 4.0 | 57.4% | 58.2% | 57.9% | 66.4% | ||
| ML engineering | PostTrainBench | 45.3% | 44.3% | 40.2% | 49.3% | |
| Science and math | Terminal-Bench Science 0.1 | 57.6% | 68.1% | 52.6% | 63.3% | |
| LABBench 2 | 88.8% | 85.4% | 68.6% | 73.1% | ||
| RiemannBench | 76.0% | 72.0% | 65.6% | 69.6% | ||
| Long context | GraphWalks | Up to 128k, BFS (F1) | 99.7% | 98.7% | 91.4% | 90.6% |
| 256k to 1M, BFS (F1) | 84.2% | 71.8% | 65.0% | 66.8% | ||
| Computer use | Agent's Last Exam | Pass rate | 39.5% | 34.2% | — | 38.2% |
| OSWorld-2.0 | Offline subset Partial score | 69.2% | 72.6% | — | — | |
| Multimodal understanding | Chartography | 71.6% | 71.0% | 46.2% | 66.3% | |
| LVBench | 91.7% | 87.5% | 79.7% | 83.7% | ||
| Cybersecurity | CWE-bench v1 | 68.0% | 68.0% | 58.0% | 67.0% |
Methodology: deepmind.google/models/evals-methodology/gemini-4-argon
Safety
Strengthening frontier safeguards before broad availability
Defending against misuse
Argon is designed to refuse harmful requests while preserving legitimate, dual-use scientific research, as per our Frontier Safety Framework.
Defending against prompt injection attacks
Argon is our most resilient model yet against indirect prompt injections.
Monitoring for misalignment
We are deploying misalignment mitigations that monitor Argon’s chain-of-thought and actions and stop execution when necessary.
Hardening systems
We are hardening our sandboxed environments by isolating and sealing them before high-risk training or evaluations begin.
Built with Argon
Get started
Try Gemini 4 Argon
Gemini
Supercharge your creativity and productivity
AI Mode
Ask whatever's on your mind to get an AI powered response
Google Antigravity
Our agent-first development platform that allows anyone to be a builder
Google AI Studio
The fastest path from prompt to production
Gemini API
Get started building with cutting-edge models
Gemini platform
Build, deploy, and govern enterprise-ready agents and apps
Android Studio
The official IDE for Android app development
The Gemini Ecosystem
Slide 1 of 4