Gemini 4 Argon Ships $2 Input, 1M Token Output

Google's Gemini 4 Argon, announced September 30, 2026, has begun rolling out to higher-tier Pro and Ultra subscribers in the Gemini app, AI mode in Search, and Ultra subscribers on Antigravity, according to The Verge's Robert Hart. Google first restricted access to trusted cyber defenders through its Fairwind Program and says it remains engaged in the U.S. government's voluntary pre-release model access process while it expands availability.

Google prices the model at an introductory $2 per million input tokens and $10 per million output tokens, with cached input tokens at 95% off the input price. Google describes the output limit as industry-leading at 1 million tokens, up from 64,000 tokens, or more than 15x.

Reported benchmarks

Google reports Argon sets a new state of the art on DeepSWE v1.1 with 77.9%, ranks first on Zapier's AutomationBench with 51.3%, and scores 91.7% on LVBench for long-video understanding. Google also says Argon leads the Vals Index, Vals Finance Agent v2, and Harvey's Legal Agent Benchmark. On cybersecurity, it ties for first on CWE-bench v1 at 68%. The Verge's Robert Hart reports that Google claims Argon outperforms GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 across coding, cybersecurity, science, finance, and law; those comparisons are not independently verified.

Inside Google, Argon agents migrated C/C++ codebases to Rust at up to 800K+ lines for the Fuchsia Zircon kernel. For libgav1, agents replaced 32K lines of SIMD code with safe Rust that vectorizes automatically, producing a memory-safe decoder that Google says runs 2.7x faster than the existing Rust port. A separate fleet-profiling deployment freed over 300 TiB of memory, with 500 TiB to 1 PiB in projected total savings, and Argon beat a published quantum algorithmic baseline by 40% in minutes, Google reports.

Pricing and competitive position

According to AI Mastery's pricing data, GPT-6 Astra and Claude Fable 5.1 each cost $10 input / $50 output per million tokens — five times Argon's introductory input price and five times its output price. Claude Sonnet 5 matches Argon at $2 input / $10 output, while Gemini 3.1 Pro is $2 / $12.

Prices for non-Argon models from AI Mastery's pricing data as of 2026-09-14. Argon pricing from Google's announcement; introductory period ends late November per The Verge.

The Verge reports that when the introductory period ends in late November, pricing will double and Argon will remain significantly cheaper than frontier rivals on a per-token basis.

AI Mastery analysis

The pricing is the immediate decision point, but the 1 million-token output limit is the architectural differentiator. As we have written in our analysis of constrained inference and generation cost, token headroom changes which agentic workflows are feasible in a single trajectory, not just how much output can be buffered. A sustained 1M output limit at production latency would make Argon structurally different from similarly priced models such as Claude Sonnet 5.

Availability remains the caveat. Argon is not open to API developers at launch, and The Verge says it is not clear when or if Plus subscribers or free users will get access. For engineering teams evaluating models this quarter, Argon is a pricing and capability anchor rather than a deployable option, and a model that cannot be tested against real workloads does not change a deployment calculus. That aligns with our reporting on production AI failures rooted in architecture, not model intelligence.

The cybersecurity-first sequencing is also deliberate: Google is releasing Argon without cyber guardrails to trusted defenders, while the broader model remains gated. It is a rollout designed to prove value in vulnerability discovery and remediation before exposing more general capabilities.

Sources

• Primary: Introducing Gemini 4 Argon
• Reported by: The Verge AI: Google finally releases Gemini 4 Argon
• Competitor prices: AI Mastery token calculator, figures as of 2026-09-14