DeepSeek V4 Pro Goes Head-to-Head with GPT-5.3 Codex High
DeepSeek V4 Pro vs GPT-5.3 Codex High: coding & Agent test results, winner revealed.
“AI Disruption” Publication 9700 Subscriptions 20% Discount Offer Link.
I ran a quick test on DeepSeek V4 Pro’s real-world coding and Agent capabilities, comparing it against GPT-5.3 Codex High. Overall result: GPT-5.3 Codex High > DeepSeek V4 Pro.
These test results are for reference only.
This test used DeepSeek V4 Pro (accessed via Claude Code) and GPT-5.3 Codex High. The evaluation model was GPT-5.5 Thinking.
The entire test consisted of two rounds.
Round 1: Algorithm problem — Implement LRU Cache in TypeScript.
LRU Cache Results:
Round 2: A more realistic engineering Agent task — Implement a local Markdown article analysis CLI tool in TypeScript.
Markdown CLI Tool Results:
Since DeepSeek released its API at a 2.5x discount last night, the entire test only cost ￥1.4.
OpenAI has been pushing frontier models at roughly a monthly pace: GPT-5.3 Codex launched on Feb 24, GPT-5.4 on March 5, and GPT-5.5 on April 23. Therefore, comparing GPT-5.3 Codex High with V4 Pro is a reasonable choice.
