At this point I've become really efficient in using AI for coding so I've tested DeepSeek yesterday and today for coding.

It's faster, better than GPT-4o at every possible situation I've tested.

At one-off prompts, it's even better than Claude Sonnet, which is my go-to model for coding. But it sometimes "over-engineers" solutions (like o1 pro). Also Sonnet is still better at remembering context + has Artifacts UI (obv this last part isn't relevant if you're using Cursor or an API).

DeepSeek is also way cheaper (even cheaper than Gemini 1.5 Flash) with higher rate limits.

🚀 Introducing DeepSeek-V3!

Biggest leap forward yet:
⚡ 60 tokens/second (3x faster than V2!)
💪 Enhanced capabilities
🛠 API compatibility intact
🌍 Fully open-source models &amp; papers

🐋 1/n https://t.co/p1dV9gJ2Sd

DeepSeek

At this point I've become really efficient in using AI for coding so I've tested DeepSeek yesterday and today for coding.

It's faster, better than GPT-4o at every possible situation I've tested.

At one-off prompts, it's even better than Claude Sonnet, which is my go-to model for coding. But it sometimes "over-engineers" solutions (like o1 pro). Also Sonnet is still better at remembering context + has Artifacts UI (obv this last part isn't relevant if you're using Cursor or an API).

DeepSeek is also way cheaper (even cheaper than Gemini 1.5 Flash) with higher rate limits.

https://x.com/deepseek_ai/status/1872242657348710721?t=HtvTimYthpItUC6sWZfIKg&s=19

microsub.

https://chainleft.art

Dad, data scientist, blockchain native artist.

Power consolidation=evil.