8 de outubro de 2026 · IA e LLMs · 4 min de leitura
We wrote about Gemini 3.7 Flash's price doubling back in September — $0.75/$3.75 per million tokens through December 31, 2026, then $1.50/$7.50 from January 1, 2027. Since then, Google shipped 3.8 Flash, and it's reasonable to assume a newer model got a cleaner rate card. It didn't. I pulled the current Gemini API pricing docs directly and checked every Flash generation Google currently sells: 3.6 Flash, 3.7 Flash and 3.8 Flash all carry the exact same intro-rate clause, on the exact same date. If your plan was "move to the newest Flash before January and dodge the hike," that plan doesn't work — there's nowhere in the current Flash lineup left to move to.
| Modelo | Agora, até 31 de dezembro de 2026 | A partir de 1º de janeiro de 2027 | Mudar |
|---|---|---|---|
| Gemini 3.8 Flash | $0.75 / $3.75 | $1.50 / $7.50 | 2.00x |
| Gemini 3.7 Flash | $0.75 / $3.75 | $1.50 / $7.50 | 2.00x |
| Gemini 3.6 Flash | $0.75 / $3.75 | $1.50 / $7.50 | 2.00x |
| Gemini 3.5 Flash (sem alteração agendada) | $1.50 / $9.00 | $1.50 / $9.00 | nenhuma |
| Gemini 3.5 Flash-Lite | $0.30 / $2.50 | $0.30 / $2.50 | nenhuma |
| Gêmeos 2.5 Flash (geração anterior) | $0.30 / $2.50 | $0.30 / $2.50 | nenhuma |
| Gemini 2.5 Flash-Lite | $0.10 / $0.40 | $0.10 / $0.40 | nenhuma |
Three model names, three separate launch dates, one identical price line. That's not a coincidence of similar models landing on similar numbers — it's the same promotional structure applied to an entire model family: cheap intro rate to pull in adoption, doubling on the same January 1 date regardless of which specific Flash build you picked. Google's docs don't call out 3.6 or 3.8 by name the way the 3.7 announcement got attention — the clause just sits under the same rate card line for all three.
Um aplicativo de chatbot que executa 300 MILHÕES de tokens de entrada e 60 MILHÕES de tokens de saída por mês — um bot de suporte de tamanho médio, não um caso de borda. Preços em qualquer modelo Flash de geração atual que esteja em execução:
| Entrada (300M) | Saída (60M) | Total mensal | |
|---|---|---|---|
| Agora (3,6 / 3,7 / 3,8 Flash, até 31 de dezembro) | $225.00 | $225.00 | $450.00 |
| A partir de 1º de janeiro de 2027 (mesmo modelo, qualquer um dos três) | $450.00 | $450.00 | $900.00 |
| Gemini 3.5 Flash-Lite, a qualquer momento (sem caminhada) | $90.00 | $150.00 | $240.00 |
O salto de $ 450 é idêntico, não importa qual das três gerações atuais do Flash o bot esteja executando — não há movimento de "atualizar para esquivar" disponível dentro do próprio nível do Flash. A única saída real é uma mudança de modalidade, não um aumento de versão: o Flash-Lite já executa essa carga de trabalho exata por $ 240/mês hoje, antes e depois de janeiro, porque nunca fez parte do preço promocional.
Don't spend engineering time migrating between 3.6, 3.7 and 3.8 Flash for cost reasons — they're the same bill, just with different model weights attached. If the January number matters to your budget, the decision that actually changes it is Flash vs. Flash-Lite (or staying on 2.5 Flash), decided on an eval set, not a model-version number. Run your real traffic through Flash-Lite before December 31 and measure whether answer quality holds — if it does, the $210/month difference in the example above is free money; if it doesn't, you now know to budget for the doubled rate with your eyes open instead of discovering it on the January invoice.
Novo nos preços de LLM baseados no uso? Comece com o guias gratuitos de custo de API.
Implemente você mesmo: DigitalOcean – crédito grátis de $ 200 ↗ · VPS Hostinger ↗
Pricing verified against Google's official Gemini API pricing docs (ai.google.dev/gemini-api/docs/pricing), checked 2026-10-08. Reference estimates using Google's published pay-as-you-go rates — Vertex AI pricing, enterprise agreements and future rate changes vary; confirm current pricing at the source before budgeting. Workload figures (token counts) are illustrative scenarios computed against real published per-token rates, not vendor-supplied numbers.