A fierce competition is unfolding in the artificial intelligence programming assistant sector, with recent data indicating that OpenAI's coding assistant Codex has surpassed Anthropic's Claude Code in several key performance areas.
According to data from startup Modu analyzing over 300,000 code pull requests on October 9th, developers approved code generated by Codex at a rate of 74.3%, slightly exceeding Claude Code's 73.7% approval rate.
This performance shift has gained market validation, with developers on social platforms including X and Reddit noting significant improvements in Codex's capabilities over the past month.
The turnaround is largely attributed to OpenAI's GPT-5-Codex model released last month. According to Modu co-founder Brexton Pham, the previous version of OpenAI's Codex achieved only a 69% code success rate before the new model launch. The updated model has become the key factor enabling OpenAI to regain ground in this competitive race.
For Anthropic, winning the coding competition appears particularly significant. According to reports, the company's revenue primarily comes from selling its AI models through APIs to clients like Microsoft, with coding technology serving as a core revenue driver.
In contrast, OpenAI, which operates the commercial powerhouse ChatGPT, shows less revenue dependence on coding services. However, company leadership views coding as a "critical component" for developing artificial general intelligence (AGI) and doubled down on investments last year to enhance their model's coding capabilities.
**Performance Leads, But User Adoption Remains Challenge**
Despite achieving a marginal advantage in code generation quality, Codex continues to face substantial challenges in attracting developer adoption. Data shows that among code requests ultimately merged by developers, Codex-generated code accounts for 24.9% compared to Claude Code's 32.1%.
However, Brexton Pham noted that Codex's market share has improved by five percentage points since the new model launch, indicating positive momentum.
Yet from another perspective, the gap remains substantial. According to NPM software registry data, Claude Code currently registers over 5 million weekly downloads compared to Codex's 190,000.
This suggests OpenAI faces a considerable journey to convert performance advantages into market dominance, with changing developer usage habits representing a primary obstacle.
**Intense Market Competition, Cost Not Primary Factor**
Interestingly, cost appears not to be developers' primary consideration when selecting tools. Brexton Pham indicated that despite Codex being cheaper than Claude Code, developers currently show willingness to pay premium prices for high-performance products, as they generally expect long-term cost reductions. For enterprise decision-makers, investing in AI programming assistants to boost existing engineer efficiency costs significantly less than hiring additional engineering talent.
Furthermore, the AI coding market extends beyond these two giants. Modu's data reveals several emerging competitors. For instance, Sourcegraph's Amp agent achieved a 76.8% approval rate, making it the most developer-approved coding assistant. Brexton Pham describes it as a "luxury premium" product in the market - higher priced but superior performing.
Meanwhile, Google's Gemini CLI leverages its consistent low-price strategy to become the most affordable coding assistant currently available. Notably, Modu's benchmarking has not yet included major products like GitHub Copilot and Google's Jules agent, indicating the complete competitive landscape remains to be fully revealed.