Gemini 3 Flash (Preview) is built to deliver frontier-ish capability at Flash speed—great for agentic workflows, coding helpers, document intelligence, and high-volume production apps where latency and cost matter as much as quality. It’s rolling out across Gemini API (AI Studio), Vertex AI, and other Google dev surfaces, and it’s also becoming the default in parts of the consumer Gemini experience.
Architecture: Transformer-based generative multimodal LLM
Context Length: Up to 1M tokens input / 64K tokens output
Capabilities: Text + multimodal understanding (images, audio, video, PDFs), structured outputs, tool/function calling, agent-friendly behaviors.
Knowledge cutoff: Jan 2025
Inference features: Reasoning (“thinking”) is supported, and output pricing explicitly includes thinking tokens in Gemini API pricing.
Speed / throughput: Independent pre-release testing reported ~218 output tokens/sec for Gemini 3 Flash Preview (fast enough for real-time-ish agent loops and high-QPS backends).
Quality vs prior Flash: Google/DeepMind positions Flash as meaningfully stronger than previous Flash generations for difficult extraction and doc tasks.
Gemini 3 Flash is a strong default when you need:
Gemini 3 Pro is the “reasoning-first” flagship, while Flash targets similar agentic/coding workflows but optimized for cost/latency.
Gemini 3 Flash is positioned as a major upgrade (replacing the prior default Flash), aiming for more detailed answers while staying fast.
GPT-5.2 is a reasoning/coding-first flagship that’s typically chosen when you want the most dependable multi-step analysis, strong code correctness, and consistent “final answer” polish in complex, real-world workflows. Gemini 3 Flash (Preview) targets many of the same agentic/coding and document-intelligence use cases, but it’s tuned primarily for low latency and high throughput, making it a more “speed-first” default when responsiveness matters. The other big practical difference is context behavior: Flash emphasizes very long input context for feeding large corpora or long documents, while GPT-5.2 emphasizes high-quality long-form reasoning and large outputs for deeply structured work.
Gemini 3 Flash applies policy-based safety filtering and can block or stop generations for restricted categories. Guardrails can feel stricter on edge-case prompts (especially around sensitive topics), and long-context or high “thinking” settings may increase latency and token usage, so production apps usually need fallback prompts and clear UX for refusals.
Measured on the AI/ML API gateway across live traffic. Figures are medians unless stated. Last updated 2026-10-12.
Gemini 3 Flash Preview is documented as suited for coding, complex analysis, and quick answers in interactive apps. It is also described as useful for complex video analysis, data extraction, and visual Q&A.
It is documented as appropriate for agentic workflows and tool use, including multi-turn chat and collaborative coding tasks. It is also presented as useful for document analysis and deepfake detection.
import requests r = requests.post( "https://api.aimlapi.com/v1/chat/completions", headers={"Authorization": "Bearer " + AIMLAPI_KEY}, json={ "model": "google/gemini-3-flash-preview", "messages": [ { "role": "user", "content": "Hello!" } ] }, ) print(r.json())
const r = await fetch("https://api.aimlapi.com/v1/chat/completions", { method: "POST", headers: { Authorization: `Bearer ${process.env.AIMLAPI_KEY}`, "Content-Type": "application/json", }, body: JSON.stringify({ "model": "google/gemini-3-flash-preview", "messages": [ { "role": "user", "content": "Hello!" } ] }), }); console.log(await r.json());
curl -X POST https://api.aimlapi.com/v1/chat/completions \ -H "Authorization: Bearer $AIMLAPI_KEY" \ -H "Content-Type: application/json" \ -d '{"model":"google/gemini-3-flash-preview","messages":[{"role":"user","content":"Hello!"}]}'
OpenAI-compatible — swap the base URL and it works with your existing SDK.
| Type | Price |
|---|---|
| Input | |
| Output | |
| Cached input |
| Benchmark | Score | What it measures | Source | Retrieved |
|---|---|---|---|---|
| Intelligence | 26.3 | Composite score across standardised reasoning, knowledge and problem-solving evaluations, measured independently by Artificial Analysis | Source | September 12, 2026 |
| Math | 97 | Composite score across standardised mathematics evaluations, measured independently by Artificial Analysis | Source | September 12, 2026 |
| GPQA Diamond | 90.4% | Google-proof graduate science questions (hardest subset) | Source | September 22, 2026 |
| Humanity's Last Exam | 33.7% | Expert-level questions across many domains | Source | September 22, 2026 |
| SWE-bench Verified | 78% | Resolving verified real GitHub issues | Source | September 22, 2026 |
| Model | Input | Output | Context | Best for |
|---|---|---|---|---|
Gemini 3 Flash This page | Coding + agents | |||
| Reasoning + agents | ||||
| Balanced coding + agents | ||||
| Long-context, multimodal & agentic workflows | ||||
| Reasoning + agents |
Gemini 3 Flash has a 1,000,000 tokens context window and can return up to 65,536 tokens.
Gemini 3 Flash takes image, text as input and returns text.
Use google/gemini-3-flash-preview as the model id. Requests go to https://api.aimlapi.com/v1/chat/completions.
Gemini 3 Flash became available on December 17, 2025.
Gemini 3 Flash is priced at input $0.65 / 1M tokens, output $3.9 / 1M tokens, cached input $0.65 / 1M tokens.
Yes, Gemini 3 Flash can stream responses as they are generated.
Yes, Gemini 3 Flash accepts image input alongside text.
Gemini 3 Flash was built by Google.
No. It accepts image and text input but only produces text output.
