Google's new flagship AI model arrived this week with a price tag that has a countdown on it. Gemini 4 Argon, released to developers and businesses worldwide, starts at $2 per million input tokens and $10 per million output tokens — but only until November 30. After that, every tier doubles. If you pay for API calls, that deadline matters more than any benchmark in the launch announcement.
We checked the pricing tier by tier against Google's published rates and worked through three realistic monthly bill scenarios — these are calculations from the published rates, not live usage tests.
What Gemini 4 Argon actually is
Google describes Argon as a "deep work" model for complex, multi-step tasks in coding, finance, law, cybersecurity, and science — the model you hand a long job to, not a one-line question. A multi-file coding refactor, a contract review spanning dozens of pages, a research task needing several rounds of reasoning: that is the work Argon is priced for. It is the latest step in Google's fast-moving agent push — we recently separated hype from reality in Google's Gemini agent for business.
Availability is where the fine print starts. Argon is available through the paid Gemini API and Google AI Studio. Google AI Pro and Ultra subscribers can use it in the Gemini app and in AI Mode in Search. The Antigravity coding environment is limited to Ultra subscribers. What you will not find is a free tier: every API call costs money from the first token.
Early customers include Deloitte, LexisNexis, Ambience Healthcare, and Isomorphic Labs — consulting, legal data, healthcare, and drug discovery, matching the "deep work" pitch. Google claims teams using Argon in Antigravity increased agentic code submissions by more than 35% while cutting rollbacks by nearly a third. Treat that as a vendor claim — but it shows where Google thinks the model earns its price.
The 5 pricing facts that decide your bill
Fact 1: Launch pricing is $2 in, $10 out. The Standard tier charges $2 per million input tokens and $10 per million output tokens. Because Argon is a reasoning model, output is where the money goes.
Fact 2: Every price doubles on December 1. The $2/$10 rate is a launch window, not the real price. From December 1, Standard becomes $4 per million input tokens and $20 per million output tokens. Any budget or client proposal written against launch pricing will be wrong within weeks.
Fact 3: There are three tiers, and the cheap one is half price. Batch and Flex trade speed for price at $1 input and $5 output per million tokens — exactly half of Standard, doubling to $2/$10 in December. Priority costs more for faster responses: $3.60 input and $18 output, doubling to $7.20/$36.
Fact 4: Cached input is nearly free, and thinking tokens count as output. Cached input — text the model has already processed, like a repeated system prompt — costs just $0.10 per million tokens at launch ($0.20 from December). But output prices include thinking tokens, the model's internal reasoning. Long reasoning chains on simple questions are the easiest way to overpay.
Fact 5: Access is gated by subscription. No free tier on the API. App access needs a Pro or Ultra subscription. Antigravity needs Ultra. For freelancers, the cheapest door in is a subscription you may already pay for — not the API.
What your bill actually looks like: 3 real scenarios
Here is what three realistic monthly workloads cost at launch pricing, at December pricing, and on the Batch tier — ClientLab calculations from Google's published rates.
Side project: 2 million input tokens, 1 million output tokens. Launch: $14 a month. December: $28. Batch: $7 now, $14 in December. At this scale, Argon is a rounding error — test freely.
Small SaaS product: 20 million input tokens, 10 million output tokens. Launch: $140 a month. December: $280. Batch: $70 now, $140 in December. That is $1,680 a year extra at Standard rates if usage stays flat.
Agency or heavy user: 100 million input tokens, 50 million output tokens. Launch: $700 a month. December: $1,400. Batch: $350 now, $700 in December. Batch saves $8,400 a year against Standard at December prices.
One more lever: caching. If our small SaaS product repeats half its input — the same instructions and reference documents in every call — then at launch, 10 million fresh input tokens cost $20 and 10 million cached tokens cost $1, so the input line falls from $40 to $21. From December, the same split costs $42 instead of $80.
Who should switch now — and who should wait
Test now: developers with long, multi-step coding work; teams already building on Google's AI tools; anyone whose workload fits Batch and Flex — bulk jobs that can run slower. November 30 is your deadline to learn, not to commit.
Wait: casual users, who should use the Pro or Ultra app access rather than paying API rates. Anyone whose workload already runs fine on Gemini 3.8 Flash, listed at $0.75 input and $3.75 output per million tokens — roughly a third of Argon's price. And anyone who cannot finish a real evaluation before November 30.
Freelancers, take note: if you quote client work that uses Argon, quote it at December pricing — $4 in, $20 out — or you will absorb the doubling yourself. For more, see our guide to how freelancers price work in 2026, and 7 ways to win better work with AI.
The launch context Google would rather skip
Argon arrives after a rocky stretch. The company announced in September that it no longer planned to release Gemini 3.5 Pro, a model its CEO had originally said would arrive in June. It also overhauled its DeepMind lab after co-founder and CEO Demis Hassabis stepped aside and several Gemini leaders left. It is a flagship launched by a team under pressure.
Two side announcements deserve attention. Google will open its Fairwind Program to hundreds of additional partners; the program gives vetted cybersecurity defenders a version of Argon without its cyber guardrails. Separately, Google will offer 10,000 scientists six months of free access to Antigravity's highest usage tier. Both moves buy adoption while everyone else faces a price that doubles in six weeks. The launch is part of a wider AI infrastructure race — see Amazon's $7B bet on AI chips.
ClientLab take
Here is our signed opinion: the doubling is the real story of this launch, not the model. A flagship model is unremarkable news in 2026; a 100% price increase with a six-week fuse is not. Use the cheap window to test Argon on your hardest real job, measured against whatever you use today. Budget every quote at December prices. If Argon earns its place at $4 and $20, it has earned its place. If it only makes sense at $2 and $10, it was never your model — it was Google's promotion.
Sources: Reuters, "Google launches flagship AI model Gemini 4 Argon for developers, businesses"; Google's published Gemini API pricing tiers.
0 comments:
Post a Comment