Gemini 3.8 Flash doubles in price on Jan 1: how to budget
Gemini 3.8 Flash costs $0.75 per million input tokens and $3.75 per million output tokens until December 31, 2026. From January 1, 2027 every rate doubles, to $1.50 and $7.50. If your agents will still run next year, budget at the new price.
gemini-3.8-flashGoogle · 1M · Stable $0.75 / $3.75 · Discount $0.1875 / $0.9375Model page →USD per million tokens
A dash means no price is recorded, not free usage. The request receipt is the final billing record.
What changes #
| Item | Through Dec 31, 2026 | From Jan 1, 2027 |
|---|---|---|
| Input | $0.75 | $1.50 |
| Output (including thinking tokens) | $3.75 | $7.50 |
| Cache read | $0.075 | $0.15 |
| Cache storage (per million tokens per hour) | $0.50 | $1.00 |
Prices are USD per million tokens. Google's pricing page labels output as "including thinking tokens", so the model's thinking is billed at the output rate. Image, audio and video input use the same input price as text.
Batch and Flex double too ($0.375 / $1.875 → $0.75 / $3.75), and so does Priority ($1.35 / $6.75 → $2.70 / $13.50). Lazu's stable lane bills at Google's standard rate; Batch and Flex are not offered.
The wording differs between Google's pages. The launch post and the what's-new page call $0.75 / $3.75 an introductory price ending December 31, 2026; the pricing page just lists the two dated prices; the model card calls $1.50 / $7.50 the regular price. The numbers agree everywhere.
Google's launch post and model card also say 3.8 Flash may use more tokens to get better results. Compare the real cost of your own tasks, not only the per-token price.
What it means for an agent workload #
Take a typical agent step: 20K tokens of context in, 2K tokens out (including thinking).
| Now | From Jan 1 | |
|---|---|---|
| One step | 20K × 0.75 + 2K × 3.75 ≈ $0.0225 | ≈ $0.045 |
| 1,000 steps a day | ≈ $22.50 | ≈ $45 |
| A 30-day month | ≈ $675 | ≈ $1,350 |
If 15K of those 20K tokens hit the cache, one step is about $0.0124 now and $0.0248 next year. The higher your cache hit rate, the more you save in absolute terms after the change, so it is worth checking now that the fixed system prompt and tool definitions sit at the start of the context and stay identical between calls.
Three things to do now #
- Re-run your budget at the new price, especially any annual quote already sent to finance or a customer.
- Make the cache work for you. Fixed content first, changing content last.
- Try a few alternatives on your own tasks. Models differ a lot on any given task, and price is only part of it. On Lazu the same key works for all of them; change the model name:
| Model | Input / output (USD per million tokens) | Note |
|---|---|---|
| Gemini 3.7 Flash | $0.75 / $3.75 | 3.8 Flash is built on it; Google also calls its price introductory, so check Google's notice for next year |
| DeepSeek Flash | $0.1875 / $0.75 | Doubles at peak: weekdays 01:00–04:00 and 06:00–10:00 UTC |
| GPT-6 Luna | $0.10 / $0.50 | |
| Claude Haiku 5.5 | $0.10 / $0.50 | A prompt over 100K tokens prices the whole request at 5x |
Lazu stable-lane prices as of 2026-10-20; the model pages always have the current price.
What Lazu will do #
Lazu's stable lane bills at Google's official price, so from January 1, 2027 it moves to $1.50 / $7.50. The discount lane's new price will be on the model page and in the changelog before the end of December.
The price applied to each call is on its request receipt in the console.
A minimal request #
from openai import OpenAI
client = OpenAI(base_url="https://api.lazu.ai/v1", api_key="YOUR_LAZU_KEY")
reply = client.chat.completions.create(
model="gemini-3.8-flash",
messages=[{"role": "user", "content": "Explain prompt caching in one sentence."}],
)
print(reply.choices[0].message.content)
The native Gemini endpoint works too: POST https://api.lazu.ai/v1beta/models/gemini-3.8-flash:generateContent.