Gemini 4 Argon

On October 13, 2026, Google Cloud made Gemini 4 Argon generally available, positioning it as the intelligence layer behind the new Gemini Enterprise agent experience. The model is trained to maintain context across long, multi-step tasks with domain expertise, and is aimed at complex software builds, financial and legal work, and cybersecurity tools like CodeMender. It reads long mixed-format documents and produces structured, source-grounded outputs. Argon is designed to respect system boundaries, report negative findings honestly, and pause before irreversible actions on long-running tasks. Early customers using the model told Google it excels at turning large, unstructured inputs into cohesive outputs. The rollout is framed as a hand-off of deep, multi-step work that previously took hours or days.

Reuters reported on October 13, 2026, that Google released Gemini 4 Argon to developers and businesses worldwide, positioning it for complex multi-step work in coding, finance, law, cybersecurity, and science. API launch pricing starts at $2 per million input tokens and $10 per million output tokens, with rates scheduled to double after November 30. Early testers include Deloitte, LexisNexis, Ambience Healthcare, and Isomorphic Labs. Teams using Argon in Google's Antigravity coding tool increased agentic code submissions by more than 35% while cutting code rollbacks by nearly a third. Google AI Pro and Ultra subscribers can access the model through the Gemini app and AI Mode in Search, with the Antigravity coding tool limited to Ultra subscribers. The Fairwind Program is being expanded to hundreds of additional vetted cybersecurity partners.

Google DeepMind announced general availability of Gemini 4 Argon on October 13, 2026, in a post by Koray Kavukcuoglu, Chief AI Architect and SVP at Google DeepMind. The model targets deep, multi-step work in coding, finance, legal, cybersecurity, and science, and is reported to rank first on the Vals Index measuring performance in critical economic sectors. Engineering teams inside Google used Argon for large-scale refactoring, including untangling Chrome's browser-feature code across hundreds of files in a 40-million-line codebase, completing months of work in weeks. Internal teams reported more than a 35% increase in agentic code submissions and roughly a third fewer code rollbacks when using Argon in Antigravity.

Google announced Gemini 4 Argon on September 30, 2026, describing it as a frontier model built for sustained reasoning across long-horizon workflows in software engineering, legal and finance knowledge work, and cybersecurity defense. The model supports an industry-leading 1 million token context for deep, multi-step problem solving. Initial access is limited to trusted cyber defenders through the Fairwind Program, with broader release pending safety testing. Launch API pricing is set at $2 per million input tokens and $10 per million output tokens, with cached input tokens priced 95% off. Thousands of internal Googlers are already using Argon for specialized coding, deeper research, and writing tasks. Google is participating in the U.S. government's voluntary pre-release model access process while expanding access to developers, enterprises, and consumers.

Vals AI lists Gemini 4 Argon as released September 30, 2026 by Google, with a 1M context window, 262k max output tokens, and token costs of $4.00 input and $20.00 output per million. On the Vals Index it ranks 1 of 48 models with 68.90% accuracy, and its strongest result is 1 of 78 on the Finance Agent (v2) benchmark. The model supports text, image, video, and file inputs. Argon places 2 of 77 on Code Migration at 68.17% and 3 of 47 on CyberBench v1.1 at 77.86%, alongside mid-pack results on Legal Research Bench (54.81%) and ProofBench v1.1 (99.00%). The model also records 87.43% on MedScribe and 75.24% on EMB, giving an independent third-party view of strengths and gaps versus frontier claims.

Google Cloud's developer guide for Gemini 4 Argon documents a unified 1,048,576-token limit covering both input and output, including thinking and function calls. Long decodes extend generation up to 1,048,576 tokens minus input on GenerateContent, with the Google Gen AI SDK handling continuation tokens automatically. Per-request output caps are 65,536 tokens, or 32,768 for non-streaming direct HTTP. Thinking levels include LOW, MEDIUM (default on gemini-4-argon), and HIGH (default on gemini-4-argon-cyber-permissive). The guide covers quickstart, structured output, mandatory API rules, and a migration path from Gemini 3.1 Pro. Argon is the first model in the Gemini 4 family and targets deep work in coding, finance, legal, cybersecurity, and science.

Google DeepMind's Gemini 4 Argon Model Card, published October 13, 2026, documents the model's technical foundation and intended use. Argon is a large-scale sparse mixture-of-experts transformer with native multimodal support for text, vision, and audio inputs. It accepts text, images, audio, and video, producing text outputs with a combined input-output context window of up to 1 million tokens. The card confirms Argon is not a modification or fine-tune of any prior model, marking a fresh frontier release. Pre-training used a large, diverse, multimodal corpus including web documents, code, images, audio, and video, with post-training adding instruction tuning, reinforcement learning, and human-preference data including multi-step reasoning and theorem-proving examples. The card covers known limitations, safety performance, and distribution policy.