AI coding agents
Sep 4, 2026

Codex Finishes Austin Griffith’s AI Security Race, and DeepSeek Comes Close

Ten AI coding agents were tested on the same 12 Solidity challenges on Thursday, and only one model completed the course. The race was organized by Austin Griffith, who works on developer onboarding and tooling at the Ethereum Foundation and founded BuidlGuidl, and it used challenges originally built for human developers at Ethereum Devcon conferences. OpenAI’s Codex, running GPT-5.5, took all three finishing spots. The medium reasoning setting was the fastest, finishing all 12 flags in 40 minutes and 7 seconds, while the extra-high setting finished last among the three and used nearly a third more tokens. DeepSeek V4 Pro, an open-weight Chinese model, captured 11 of 12 flags for $1.45 in compute. Anthropic’s Claude Opus 4.8 solved 10 flags and cost $7.61. Other open-weight Chinese models lagged further behind, with GLM 5.3 taking six flags and Kimi K3 and Qwen taking two each. BuidlGuidl said the exercise was a transparent single-run evaluation, not a universal model ranking. Griffith’s stated point was simple: new model releases need better evals, and builders should be able to run their own.

1560
Codex Finishes Austin Griffith’s AI Security Race, and DeepSeek Comes Close