Google Releases Gemini 4 Argon API With 1M Token Output Limit

MarkTechPost’s developer-focused writeup on Gemini 4 Argon’s API availability highlights a sparse MoE frontier model for long multi-step work in coding, finance, legal, science, and cybersecurity. After a September 30 preview for cyber defenders, Argon reaches the paid Gemini API at an introductory $2 input / $10 output per million tokens, rising to $4 / $20 after November 30, 2026.

The standout limit is output length: up to 1 million tokens in a single response within a combined 1M input-plus-output window, with a documented continuation protocol for long decodes. Google’s comparisons claim top marks on several knowledge-work and long-context rows, including a #1 Vals Index score of 68.90%, while trailing rivals on some agentic coding benches such as FrontierSWE v2 and Terminal-Bench 4.0.

Access spans the paid Gemini API, AI Studio, enterprise surfaces, Pro/Ultra consumer apps, and the Fairwind Program for cyber defenders.

A million-token output ceiling is a product statement: Google wants agents that can rewrite large codebases, emit long formal artifacts, and keep going without duct-taped continuation hacks—though it still documents a continuation protocol because reality is messy.

Price is the other half. Intro rates at roughly a fifth of some rival list prices will pull experiments onto Argon whether or not every benchmark row is gospel. After November 30 the gap narrows; builders who architect around cheap long outputs need an exit plan.

We’re not swallowing the entire scoreboard. Google’s table mixes self-computed and third-party numbers, and Argon loses on several SWE/terminal benches that matter for agentic coding. Hallucinations and timeouts are acknowledged on the model card. Under the Frontier Safety Framework, Google says it cannot rule out critical capability levels in biology and cybersecurity and is deploying safeguards it believes lower the risk.

Net for us: Argon is a credible re-entry for Google at the frontier, especially for long-horizon professional work and defender tooling—if teams validate on their own harnesses instead of the launch chart.

Frontier labs are competing on long-context agents, professional workflow benchmarks, and API price/performance. Google’s Argon release emphasizes extreme output length and aggressive introductory pricing.