Why ChatGPT Uses SSE Instead of WebSockets

🎯 The Question

"When ChatGPT streams tokens to your screen in real time, why does OpenAI use Server-Sent Events (SSE) instead of bidirectional WebSockets?"

---

⚡ 30-Second Elevator Pitch

WebSockets are designed for bidirectional, full-duplex communication (like multiplayer games or live chat rooms where both client and server constantly send messages simultaneously).

LLM token streaming is strictly unidirectional:

1. The user sends 1 prompt to the server (standard HTTP POST).
2. The AI model streams hundreds of tokens back sequentially over time.

Why SSE is superior for LLM streaming:

• Runs over standard HTTP/2: Free HTTP/2 connection multiplexing, header compression, and zero custom TCP upgrade handshakes.
• Built-in Browser Reconnection: Native EventSource automatically reconnects and resumes streams if Wi-Fi drops.
• Firewall & Proxy Friendly: Uses standard HTTP/HTTPS (text/event-stream), avoiding WebSocket proxy blocking in corporate enterprise firewalls.

---

🧠 Under-the-Hood: SSE Stream vs. WebSocket Protocol

---

🔬 How SSE Works Over HTTP

The server simply responds with standard HTTP headers:

HTTP/1.1 200 OK
Content-Type: text/event-stream
Cache-Control: no-cache
Connection: keep-alive

data: {"token": "Hello"}

data: {"token": " world"}

data: [DONE]

The browser receives tokens as continuous text chunks without parsing custom binary framing protocols.

---

📌 Comparison Matrix: Server-Sent Events (SSE) vs. WebSockets

Dimension	Server-Sent Events (SSE)	WebSockets
Communication Flow	Unidirectional (Server $o$ Client)	Bidirectional (Client $\leftrightarrow$ Server)
Protocol Foundation	Standard HTTP/1.1 or HTTP/2	Custom ws:// / wss:// TCP protocol
Auto-Reconnection	✅ Built-in by browser specification	❌ Requires custom JavaScript heartbeat logic
Corporate Firewalls	✅ Passes cleanly (Standard HTTPS 443)	⚠️ Often blocked/dropped by enterprise proxies
Ideal Workloads	LLM token streaming, stock tickers, news feeds	Multiplayer games, collaborative whiteboards

---

💡 What Interviewers Ask Next (Follow-Up Traps)

1. "What is the 6-connection limit bug with SSE in HTTP/1.1?"

   • Answer: Under HTTP/1.1, browsers restrict maximum connections per domain to 6. Opening 6 SSE streams in different tabs completely blocks all subsequent HTTP requests. Solution: Serve SSE over HTTP/2, which multiplexes hundreds of streams over a single shared TCP connection.

2. "Can the client send data over an open SSE connection?"

   • Answer: No. SSE is strictly unidirectional. If the client needs to send new data (e.g. stop generating or send a new prompt), it makes a separate standard HTTP POST request.

---

Placement & Interview Takeaway

Interview Answer: ChatGPT uses Server-Sent Events (SSE) because LLM inference is strictly unidirectional (client sends a single prompt, server streams tokens back). SSE runs over standard HTTP/2 with built-in auto-reconnection and seamless firewall traversal without the overhead of WebSocket state management.

---

📺 Video Explanation