Summary: Google’s Gemini 4 Argon, the first model in the Gemini 4 family, is now listed as a third-party model on Cloudflare Workers AI, callable with a single env.AI.run() request. Through November 30, 2026, it runs at half price — $2.00 input and $10.00 output per 1M tokens — before doubling on December 1. If you’ve been waiting to prototype with a frontier model, the discount window is the practical reason to start now.
What the listing actually gives you
Cloudflare’s model page describes Argon as Google’s frontier model for deep work on complex tasks across coding, finance, legal, cybersecurity, and science. It comes with a 1,048,576-token context window and supports Chat Completions request formats.
The integration story is the part worth your attention. You invoke it from a Worker like this:
const response = await env.AI.run(
'google/gemini-4-argon',
{
contents: [
{
parts: [
{ text: 'What are the key differences between optimistic and pessimistic concurrency control in distributed systems?' },
],
role: 'user',
},
],
},
)
console.log(response)Or via a plain REST call to the AI run endpoint:
curl https://api.cloudflare.com/client/v4/accounts/$CLOUDFLARE_ACCOUNT_ID/ai/run \
--header "Authorization: Bearer $CLOUDFLARE_API_TOKEN" \
--header "Content-Type: application/json" \
--data '{
"model": "google/gemini-4-argon",
"input": {
"contents": [
{ "parts": [ { "text": "Explain MVCC briefly" } ], "role": "user" }
]
}
}'No new SDK, no separate vendor account. If your product already lives on Workers, adding a frontier model is a string change in your model name plus a pricing line in your budget.
The pricing math that should drive your decision
The discount is temporary, and the terms are specific: through November 30, 2026, you pay $2.00 per 1M input tokens, $0.10 per 1M cached input tokens, and $10.00 per 1M output tokens. Starting December 1, 2026, those return to $4.00, $0.20, and $20.00 respectively.
Two implications for builders:
- Run your evaluation and load-testing during the window. Benchmark quality, latency, and token consumption now, when the same experiments cost half as much.
- Model your December costs before you ship. A workload that fits your budget at $10/1M output tokens may not at $20. Double your projected inference spend and see whether the unit economics still work.
The cached input rate is also worth noticing: $0.10 per 1M tokens means prompts with large, stable prefixes — system prompts, retrieved documents, long code context — get dramatically cheaper on repeat calls. If you’re structuring prompts for cache hits, this model rewards that discipline.
Where Argon fits in your stack
Argon keeps showing up across builder surfaces. I’ve covered its rollout in Google’s own client and how Cursor lets you tune its effort levels against pricing. The Cloudflare listing closes a different gap: serverless deployment. Agents and background jobs that run on Workers can now reach the same frontier model without routing through another provider.
That said, this is a third-party listing on Cloudflare’s platform, not a Cloudflare-trained model — terms and licensing follow the linked agreements on the model page, so read them before committing production traffic.
A grounded next step
If you have a deferred experiment — a document pipeline, a coding assistant feature, a research agent — the half-price window through November 30, 2026 is a low-regret time to run it. Just budget against the post-December rates, not the promo, so a discount doesn’t quietly become your baseline.
Sources
View the original post on X