Course Content
FastAPI Essentials
1 sections · 32 lessons
What are the key differences between FastAPI and Flask, and why choose FastAPI for web development?
What you need to know
The deepest difference is the interface each framework speaks to its server.
WSGI (Flask)
- One request holds one worker thread or process until it finishes
- A request that waits 4 seconds blocks that worker for 4 seconds
- No native streaming or WebSockets
ASGI (FastAPI)
- One event loop holds many requests at once
- A waiting request costs a paused coroutine, not a thread
- Streaming, SSE and WebSockets are built in
| FastAPI | Flask | |
|---|---|---|
| Concurrency | Native async/await | Sync; async views run inside a worker thread |
| Validation | Automatic from type hints (Pydantic) | Manual, or an extension such as marshmallow |
| API docs | OpenAPI and Swagger UI generated from code | Extensions (flasgger, apispec) |
| Ecosystem | Younger, smaller | Older, large, many extensions |
| Server-rendered pages | Possible, not the focus | A core strength (Jinja templates) |
Here is the same sentiment endpoint in Flask. Validation is your job:
1from flask import Flask, jsonify, request23flask_app = Flask(__name__)45@flask_app.post("/sentiment")6def sentiment():7 data = request.get_json(silent=True) or {}8 text = data.get("text")9 if not isinstance(text, str) or not text.strip():10 return jsonify(error="text must be a non-empty string"), 40011 if len(text) > 2000:12 return jsonify(error="text too long"), 40013 return jsonify(label="positive", score=0.91)In FastAPI those four checking lines become text: str = Field(min_length=1, max_length=2000) on a model, and the rule also appears in the docs.
A real-life example
A food-delivery company runs an LLM gateway in Flask under Gunicorn with 4 sync workers. Every request waits about 4 seconds on the LLM provider. Each worker can finish one request per 4 seconds, so the whole service tops out near 1 request per second. At dinner-time peak, requests queue and time out, even though the CPU is almost idle.
They could add threads (4 workers × 8 threads = 32 in flight), and that is a fair first step. Moving the gateway to FastAPI with an async HTTP client lets a single worker keep hundreds of calls in flight. The limit becomes the provider's rate limit, not the web framework. They keep their Flask admin tool as it is, because it works and is not under load.
Follow-up questions to expect
- "Doesn't Flask support async now?" — Since Flask 2.0 you can write
async defviews, but each request still occupies a worker thread while it runs, so you do not get ASGI-style concurrency. Quart is the ASGI version of Flask. - "Is FastAPI faster for CPU-heavy work?" — No. Both are Python. Framework choice matters for I/O-bound concurrency, not for number crunching.
- "How would you migrate a large Flask app?" — Gradually. Put new endpoints in FastAPI behind the same gateway, or mount the Flask app inside FastAPI with a WSGI adapter such as
a2wsgi, and move routes one at a time.