Claude Code vs Codex: Stop Picking a Side, Start Picking a Task
In January 2026, Andrej Karpathy posted a thread about going from 80 percent manual coding to 80 percent agent coding in a single month. It pulled 40,000 likes in a week, and the replies split exactly down the middle: half pointing at Claude Code, half pointing at Codex. Six months later, that split is still the live debate, and most of the content about it is trash. It is either written by…
In January 2026, a post by Andrej Karpathy gained significant attention for suggesting a shift from manual coding to agent coding. The post garnered 40,000 likes in a week, sparking a debate that continues six months later, with most content being uninformative. This article aims to provide a factual comparison between Claude Code and Codex.
The pricing asymmetry is the first factor to consider. Both tools operate on chat subscriptions, with Codex integrated into ChatGPT plans and Claude Code with Claude plans. While the surface-level pricing appears similar ($20, then $100, then $200), the specifics differ. A $20 plan for Codex allows for 15 to 80 GPT-5.5 messages per 5-hour window, 30 to 150 GPT-5.3-Codex messages, and 10 to 60 cloud tasks.
In contrast, Claude Code's $20 plan supports light usage, with Max 5x at $100, shared between claude.ai chat and Claude Code.
The true cost lies in tokens per task. A controlled test found that Claude Code used roughly 192,000 tokens at about $2.50, while Codex used 136,000 tokens at about $2.04. This results in a 1.4x token gap and a 23 percent cost gap. Claude Code's additional tokens often lead to a more decomposed architecture and an unprompted smoke test, while Codex may leave a task empty due to misconfigured tool calls.
On public benchmarks, the two model families are nearly identical. SWE-bench Verified scores of Anthropic's Opus 4.6 (78.3 percent) and OpenAI's GPT-5.1-Codex-Max (77.9 percent) show minimal difference. However, community results indicate that Claude Code may perform better in frontend and UI work due to its richer skill set and MCP ecosystem around frontend tooling. Conversely, Codex tends to excel in long-running backend tasks due to its cloud agent's ability to run unattended for extended periods and its compactness.
Ultimately, the choice between Claude Code and Codex depends on the specific task at hand, rather than loyalty to a particular company. Claude Code offers a deeper programmable harness with features such as Subagents, Skills, Hooks, and Dynamic Workflows, making it a strong choice for complex, tool-heavy tasks. Codex, on the other hand, is preferred for its cost-efficiency and performance in backend tasks.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
