Tuesday, October 13, 2026
Tech Beat
Oct 13, 2026, 4:01 PMArtificial Intelligence

Google Launches Gemini 4 Argon With 1M Token Limit and Top Benchmark Claims

Google launches Gemini 4 Argon globally, touting a 1M token limit, benchmark leads, faster coding and science work, and stronger AI safeguards for users.

Listen to this briefingAudio briefing

Summary

On October 13, 2026, Google launched Gemini 4 Argon globally through Gemini API in AI Studio and Gemini platform, Google Antigravity, Android Studio, Gemini for business, Gemini app and AI Mode in Search. Introductory pricing is $2 per million input tokens and $10 per million output tokens, with cached input 95% off. On November 30, rates double to $4 and $20. Argon has a unified 1 million token input and output limit. Google says it ranks first on Vals Index and scored 87% on ERQA.

Internal deployments lifted agentic code submissions more than 35% and cut rollbacks nearly one-third. Argon reduced a Chromium refactor across hundreds of files in its 40 million line codebase, including a 6 million line legacy folder, from months to weeks; improved Agent Development Kit session preparation 40% and cold imports 21%; accelerated xTB predictions 2.3x and conformer search 1.8x; and raised Gemma 4 E2B inference on an A100 via vLLM 2.85x, from about 37,900 to over 108,000 tokens per second without quality loss. Isomorphic Labs cut patent and PDF synthesis from weeks to hours. Wiz used Argon to remediate government exposure and a healthcare bypass risking hundreds of thousands of patient records.

Trials nearly doubled AlphaQubit throughput; cut quantum arcsin spacetime cost more than 100x; broadened Grad’s conjecture counterexamples with Professor Matt Landreman; scored 57% on Gamow Labs’ RareBench and 52.2% on LabBench; produced five active serine esterases in hours; and taught Apptronik’s Apollo to sink seven consecutive cornhole shots. Safeguards held bio task unlocks below 1%, nearly halved harmful-request violations, cut unauthorized shortcuts to 2% and prompt injection success to 0.7% across 15 attempts. Google is expanding Fairwind to hundreds more partners, offering 10,000 scientists six free months of Antigravity’s highest tier, and plans more Gemini 4 models.

Positives

  • Argon increased Google’s agentic code submissions by more than 35% while reducing code rollbacks by nearly one-third.
  • Chromium refactoring across hundreds of files took weeks instead of months, targeting crashes affecting Chrome’s billions of users.
  • Gemma 4 E2B inference reached over 108,000 tokens per second, a 2.85x gain with full token parity and no quality loss.
  • Quantum arcsin implementation cut spacetime cost by more than 100x, while AlphaQubit throughput nearly doubled.
  • Gamow Labs recorded 57% on RareBench and 52.2% on LabBench after Argon identified a chromosomal rearrangement missed for years.
  • 10,000 selected scientists can receive Antigravity’s highest usage tier free for six months.

Risks & concerns

  • Public API prices double November 30, 2026, reaching $4 per million input tokens and $20 per million output tokens.
  • Argon’s coding and scientific capabilities create dual-use risks involving exploit development and dangerous biological materials.
  • External bio red teams still unlocked less than 1% of tested biological subtasks despite Google’s layered defenses.
  • Unauthorized shortcut attempts remained at 2%, while indirect prompt injection succeeded in 0.7% of 15 adaptive attempts.
  • Free researcher access remains subject to eligibility, capacity and selection, while Fairwind access without cyber guardrails requires approval.
Primary sourceGoogle DeepMind Newshttps://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon/
Read full article
Editorial note: Tech Beat summarizes and analyzes third-party reporting. The source link is the authoritative article. This page does not reproduce the full source text.

More From The Wire

Artificial IntelligenceOct 13

Google Rolls Out Gemini 4 Argon With New Cyber, Bio and Agent Safety Controls

Artificial IntelligenceOct 13

Niobia Uses Private AI to Diagnose Million Dollar Chip and Battery Defects

Artificial IntelligenceOct 13

Flox Intelligence Seeks $10 Million to Scale Wildlife Deterrent AI