Google Cloud launches Gemini 4 Argon for long, multi-step business work
Google Cloud has made its new frontier model, Gemini 4 Argon, available to business customers. According to Google, the model was trained to keep context across long, multi-step tasks in areas such as software development, finance, legal work and cybersecurity.
- Google Cloud made Gemini 4 Argon available in Gemini for business and via API under the model ID gemini-4-argon.
- Introductory pricing is $2 per million input tokens and $10 per million output tokens through November 30, 2026, then $4 and $20.
- Google says Argon leads the Vals Index and Harvey's Legal Agent Benchmark, and cuts indirect prompt injection attack success to 0.7%.
- An unmitigated version is limited to trusted defenders in the Fairwind Program.
What Argon is built to do
Michael Gerstenhaber, VP of Product Management for Gemini at Google Cloud, wrote that the company trained Argon to work with "depth and precision". The aim is to let the model keep context over long tasks while applying domain expertise along the way. Google names complex software builds, detailed financial and legal work, and cybersecurity tools like CodeMender as target uses.
The release follows the previous week's update to Gemini Enterprise, which introduced the Gemini agent, described by Google as a universal agent for work. Google says Argon provides the model intelligence underneath it.
According to Google, Argon can read long documents that mix text, tables, charts and images and turn them into one output grounded in those sources. It is also trained to respect system boundaries, report negative findings honestly, and pause for confirmation before taking irreversible steps. The company says it tells users when it could not find something or finish a task.
Benchmarks and customer feedback
Google says Argon leads on the Vals Index, which measures economic impact across finance, coding, legal and tax work. It also reports a lead on Harvey's Legal Agent Benchmark for legal research and drafting. Niko Grupen of Harvey said the company saw substantial improvements over Gemini 3.1 Pro across its legal agent evaluations, particularly in corporate M&A, data privacy and cybersecurity, and EC/VC.
ServiceNow's Nenshad Bardoliwalla said that in early access the model delivered roughly 20% improved task success than Gemini 3.8 Flash on NowAI-Bench, with standout gains in HR and customer service. Figma's Loredana Crisan said Argon handles complex tasks reliably, from 3D games to rich animations, in Figma Make.
Google also shared results from its own engineering. Working autonomously overnight within automated testing guardrails on the open-source Agent Development Kit, Argon found optimizations now live in google/adk-python. According to Google, they cut in-memory session prep time by 40% in long sessions and speed up cold imports by 21%. Teams using Argon in Google Antigravity increased agentic code submissions by more than 35% while cutting code rollbacks by nearly a third. The Chrome team used it for an architectural refactor across hundreds of files in Chromium's 40 million-line codebase, finishing months of work in weeks, Google says.
Google adds that its AlphaEvolve agent uses Argon for gains in areas including factory scheduling, logistics, game code speed-ups and chip layout.
Security measures
Argon has cyber mitigations that block offensive tasks such as pentesting and exploitation, according to Google. Defensive work remains open through CodeMender or a customer's own security harness. Google says the model ties for first place on CWE-Bench v1 with a top score of 68% for remediating security vulnerabilities.
The company also calls Argon its most resilient model yet against indirect prompt injections. It says the model leads the Gray Swan Indirect Prompt Injection benchmark, cutting attack success across 15 adaptive attempts to 0.7%. HubX's Yunus Emre said that in the company's evaluations the model identified 100% of top-priority vulnerabilities with 0% fabricated vulnerabilities across verified claims.
A version without cyber guardrails is being released to a select set of trusted defenders. They are vetted through the Fairwind Program, a limited access initiative for government, critical infrastructure and cybersecurity partners.
Availability and pricing
Gemini 4 Argon is available in Gemini for business on the web, desktop and mobile. Developers can use it through the API in the Gemini platform under the model ID gemini-4-argon, in Google AI Studio and in Google Antigravity.
- Introductory rate through November 30, 2026: $2 per million input tokens and $10 per million output tokens, with cached input tokens priced at 95% off.
- Standard pricing after November 30: $4 per million input tokens and $20 per million output tokens.
For businesses, the pitch is handing off multi-step work that used to take hours or days. Google points teams to the model card and developer guide for technical specifications, token limits and migration instructions. The benchmark and productivity figures come from Google and its early access customers.
