What changed
Google released Gemini 4 Argon on September 30 at $2 per million input tokens and $10 per million output tokens. The company calls it a frontier model trained for coding, cybersecurity and knowledge work. Cached input costs $0.10 per million tokens. OpenAI's GPT-6.1 Sol launched the day before at the same price. Both undercut GPT-6 Astra, which charges $10 input and $50 output. Argon scored 77.9% on DeepSWE, a software engineering benchmark, ahead of Claude Opus 5.5 at 74.2% and GPT-6 Astra at 74.1%. Google is releasing Argon first to cybersecurity partners, then to paid API customers.
What you can do with it this week
Look at the tasks you skipped last month because the bill ran too high. A 50-page contract summary that cost $3 on Astra now costs $0.60 on Argon. Migrating 200 product descriptions to a new format that cost $40 now costs $8. Debugging a messy spreadsheet formula that took $5 now takes $1. If you already use ChatGPT Pro or Claude, test the same prompt on three models and compare the bill. Run the task on Argon or Sol first. If the answer works, you just cut the cost by 80%. If it does not, step up to Astra and pay the difference only when you need it.
What to keep in your own hands
Google labeled this price introductory, which means it can rise. Before you build a weekly workflow around Argon, check two things: can you move the same task to another model if the price doubles? Can you re-create the output yourself if the API shuts off for a day? Lock your spending with a monthly budget cap in the API console so a runaway loop does not drain the account. Save one finished output each week as a sample. If the model or the price changes, you have proof of what worked and a template to hand to the next tool.