9 November 2026 · API Pricing News · 5 min read
Google quietly set an expiration date on Gemini 4 Argon's launch pricing. Since the model opened up on October 13, it's been billing at $2 per million input tokens and $10 per million output tokens -- cheap enough that it undercut most of the field at launch. That rate was never permanent. On December 1, it doubles to $4/$20. I went looking for the exact numbers because "doubles" gets thrown around loosely in pricing news, and this time it's literally true, down to the decimal.
| Métrique | Now (through Nov 30) | From Dec 1 | Changement |
|---|---|---|---|
| Entrée / 1 million de jetons | $2.00 | $4.00 | +100% |
| Sortie / 1 million de jetons | $10.00 | $20.00 | +100% |
| Cached input / 1M | $0.10 | $0.20 | +100% |
| Priority tier, input/output / 1M | $3.60 / $18.00 | $7.20 / $36.00 | +100% |
Every figure in that table doubles at the same rate. That's unusual on its own -- most repricings shift input and output by different amounts, or leave caching alone to soften the headline number. Here even the Priority service tier, the one meant for latency-sensitive production traffic, moves in exact lockstep. The current rate was promotional from day one: Google's own launch terms describe it as effectively a 50% credit back on net spend against the $4/$20 list price, not a separate permanent tier. December 1 isn't a price hike so much as a discount expiring on schedule.
Skip the "per million tokens" abstraction and put a real monthly volume through it. Take a mid-size production app -- a support chatbot or internal RAG tool doing 50 million input tokens and 10 million output tokens a month, numbers in the range our own Calculateur de coût RAG sees from real inputs:
| At $2/$10 (now) | At $4/$20 (Dec 1+) | |
|---|---|---|
| Input: 50M tokens | 50 × $2 = $100 | 50 × $4 = $200 |
| Output: 10M tokens | 10 × $10 = $100 | 10 × $20 = $200 |
| Total | $200/mo | $400/mo |
$200 becomes $400. No asterisk, no hidden second multiplier buried in a footnote -- the one thing I keep finding when I check these "doubles!" headlines against the actual rate card is that the multiplier is usually softer in practice, because caching or a free tier absorbs part of it. Not here. If you're running Argon with meaningful cache hit rates, your cached-input spend doubles right along with everything else, so caching doesn't buy you any extra protection from this particular increase the way it normally would against a flat-output-only price rise.
Argon isn't the only Gemini 4 model on a promotional clock. Gemini 3.8 Flash's current rate also steps up, to $1.50 input / $7.50 output per million tokens, effective January 1, 2027. If you're budgeting a Gemini-based stack for next year, both dates matter: Argon's increase lands before your December invoice, Flash's lands on your January one. Lining those up against a calendar beats finding out from two separate invoices a month apart.
Our Gemini API Cost Calculator still models the previous generation -- 2.5 Pro, 2.5 Flash and 2.0 Flash -- not yet Argon. That's deliberate for now: adding a tier at a rate that's contractually temporary for another three weeks would mean ripping it out again the moment December 1 hits, and we'd rather ship the number that's actually going to be true in a month than the one that's true today. Once the $4/$20 steady-state rate is live, it goes in. If you're pricing a new Argon integration before then, use the table above directly rather than our calculator's current Gemini tiers, which price an older model family entirely.
New to why providers split input, output, cached and "priority" into separate meters in the first place? Start with the guides gratuits sur les coûts des API.
Déployez-le vous-même : DigitalOcean — 200 $ de crédit gratuit ↗ · VPS Hostinger ↗
Gemini 4 Argon and 3.8 Flash figures per Google's published developer pricing terms for the October 2026 launch and the announced December 1, 2026 / January 1, 2027 rate changes, cross-checked against multiple independent pricing-tracker writeups. Reference figures, not a live quote -- confirm current terms at ai.google.dev before budgeting at volume, especially this close to a scheduled change.