ChatGPT Plus vs Claude Pro vs DeepSeek API: squeezing the most work out of ~₩30,000 a month

One-line take: ChatGPT Plus for general chat, Claude 3.5 Sonnet (wired through Cursor) for messy refactors and architecture, DeepSeek API when you need volume and a lower bill.

1. What the three services actually look like at work

2. Pros and cons from the desk, not the brochure

1) Claude 3.5 Sonnet: still the default for developers

• Pros: Least hallucination when generating frontend UI or refactoring messy backend logic.
• Cons: Rate limits are tight. A few long code sessions over 4–5 hours and you get blocked.
• Workaround: Skip the Claude Pro web UI. Cursor IDE or an API-key tool is more productive.

2) DeepSeek-V3 / R1: price destruction plus open weights

• Pros: Roughly 1/20 of OpenAI’s price at similar quality. R1’s thinking pass holds up on math and awkward algorithms, close to o1.
• Cons: Official servers can wobble when traffic spikes.
• Workaround: Run a Distill model (Qwen 32B) on a local workstation, or dual-home a paid API endpoint as failover.

3. A ~₩30,000 portfolio that actually ships work

Recommended mix:
1. Cursor Pro ($20/mo) → near-unlimited Claude 3.5 Sonnet coding
2. DeepSeek API ($5 credit) → bulk parsing, scripts, summary bots
3. Local Ollama ($0) → private or sensitive data stays on the box
-----------------------------------------------------------
Spend: about $25 (~₩34,000) → a full-stack pipeline without an enterprise contract