Gemini 3.1 Pro Preview
Very-long-context multimodal model with implicit caching - savings appear without breakpoints, but capture is less predictable than an explicit scheme.
At a glance.
| Provider | |
| Family | gemini-3.1 |
| Released | 2026-02 |
| License | Proprietary |
| Context window | 1M tokens |
| Max output | 66K tokens |
| Modalities | text, image, video, audio, pdf |
| Tool calling | Yes |
| Reasoning mode | Yes |
| Caching | implicit |
| Batch discount | 50% off |
What reuse looks like here.
What you actually pay once caching works.
At a typical 55% prefix reuse, a million input tokens on Gemini 3.1 Pro Preview effectively costs $1.01 instead of $2.00 - blending to roughly $3.76 with a 25% output share. Background work drops a further 50% on the batch tier.
Estimate it for your workloadRoutes through these aliases:
Same OpenAI client, this model.
from openai import OpenAI
client = OpenAI(base_url="https://api.zumik.ai/v1", api_key="zk_live_...")
r = client.responses.create(
model="gemini-3-1-pro-preview", # or an alias like auto.best
input="Draft a fix for the failing test.",
)
print(r.usage.input_tokens_cached) # confirm reuseHow Gemini 3.1 Pro Preview stacks up.
Other options for these workloads.
Gemini 3.1 Pro Preview, answered.
How much does Gemini 3.1 Pro Preview cost?
Gemini 3.1 Pro Preview is $2.00 per million input tokens and $12.00 per million output tokens through Zumik. Cache reads are $0.20 per million, a 90% discount on input.
What is Gemini 3.1 Pro Preview's context window?
Gemini 3.1 Pro Preview supports a 1M-token context window with up to 66K output tokens.
Does Gemini 3.1 Pro Preview support prompt caching?
Yes. Google Gemini uses Implicit context caching caching. In the Zumik corpus, Gemini 3.1 Pro Preview shows a median cache capture of 82% on agent workloads.
Which Zumik aliases route to Gemini 3.1 Pro Preview?
Gemini 3.1 Pro Preview is a candidate for the auto.best, vision.balanced aliases, selected when it wins under current routing policy.
Run Gemini 3.1 Pro Preview with reuse measured.
Point an OpenAI client at Zumik and see exactly how much of this model's input you are reusing.