Announced April 22, 2026, Gemini 3 Flash brings Pro-grade reasoning to the Flash latency tier. 1M-token context, fully multimodal (text, image, audio, video), 65K max output. The default model in the Gemini app and AI Mode in Search. PhD-level reasoning on common benchmarks at a fraction of the cost of 3.1 Pro. Recommended for high-throughput agentic workflows, real-time multimodal chat, RAG and consumer applications.
Google DeepMind is the merged AI research organisation formed in April 2023 by combining Google Brain with DeepMind. Demis Hassabis leads the unit as CEO. Flash variants have been Google's high-throughput tier since Gemini 1.5 Flash (May 2024), with Gemini 2.0 Flash (December 2024), 2.5 Flash (mid-2025) and Gemini 3 Flash (April 2026) representing the progression. DeepMind's seminal papers include 'Attention Is All You Need' (2017), AlphaGo (2016), AlphaFold (2018-2021, Nobel Prize 2024) and the Gemini Technical Report.
Gemini 3 Flash was announced April 22, 2026 as the default Flash-tier model and the new default model in the Gemini app and AI Mode in Search. It is a natively multimodal Sparse MoE Transformer engineered to combine Gemini 3 Pro's reasoning quality with Flash-grade latency, efficiency and cost. Pretraining used Google's TPU v6e infrastructure on a multi-trillion-token mixture of web text, code, books, image-text pairs, audio and video frames. Post-training combined supervised fine-tuning, RLHF, RL against verifiable rewards and distillation from larger Gemini 3.1 Pro teacher models. The architecture preserves Gemini's native multimodality across text, image, audio and video, the full tool-use API and Search grounding, while running at a fraction of Pro pricing. Gemini 3 Flash is the recommended default for high-throughput agentic workflows and consumer-facing multimodal chat.
Paramètres
Undisclosed (sparse MoE, smaller and sparser than Gemini 3.1 Pro)
Pretrained on a multi-trillion-token mixture of web text, code, books, scientific papers, image-text pairs, audio and video frames. Heavily distilled from larger Gemini 3.1 Pro teacher models. Post-training uses supervised fine-tuning, RLHF and RL against verifiable rewards. Knowledge cutoff in late 2025.
Licence: Proprietary commercial license via Google AI Studio, Vertex AI and the Gemini app. Free tier available in the Gemini app and AI Mode in Search.
curl https://railwail.com/api/v1/chat/completions \
-H "Authorization: Bearer $RAILWAIL_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gemini-3-flash",
"messages": [
{
"role": "user",
"content": "Explain what a vector database is in two sentences."
}
],
"max_tokens": 1024
}'
import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["RAILWAIL_API_KEY"],
base_url="https://railwail.com/api/v1",
)
completion = client.chat.completions.create(
model="gemini-3-flash",
messages=[
{
"role": "user",
"content": "Explain what a vector database is in two sentences.",
},
],
max_tokens=1024,
)
print(completion.choices[0].message.content)
import OpenAI from "openai";
const client = new OpenAI({
apiKey: process.env.RAILWAIL_API_KEY,
baseURL: "https://railwail.com/api/v1",
});
const completion = await client.chat.completions.create({
model: "gemini-3-flash",
messages: [
{
role: "user",
content: "Explain what a vector database is in two sentences."
}
],
max_tokens: 1024
});
console.log(completion.choices[0].message.content);
// npm install railwail
import railwail from "railwail";
const rw = railwail(process.env.RAILWAIL_API_KEY);
const res = await rw.chat("gemini-3-flash", [
{ role: "user", content: "Explain what a vector database is in two sentences." },
], { max_tokens: 1024 });
console.log(res.choices[0].message.content);
