OpenAI updated Codex with three features that close the gap with Anthropic’s Claude Code: computer use that lets the agent see, click, and type inside applications; an in-app browser for pulling context during coding sessions; and automated PR review on real repositories. The updates shipped across April 2026 and are available to all ChatGPT plan tiers, including the free tier.

Independent Testing Results

The New Stack ran both agents against a real-world Python codebase (HTTPie, an open-source HTTP client) and found Codex now matches Claude Code’s capabilities for autonomous bug-fixing and code analysis. Both agents navigated the filesystem, read and wrote code, and executed validation steps without manual intervention.

The testing revealed a style difference rather than a capability gap. Codex operates literally: it does exactly what you ask, which makes it stronger at catching subtle bugs and backend logic errors but requires more precise prompting. Claude Code infers intent from loose prompts and produces cleaner frontend code, according to Softr’s comparison, but burns through usage faster.

Feature Parity, Different Trade-offs

Computer use, which launched on April 16 for macOS, lets Codex operate applications visually while the developer works on other tasks. The feature puts Codex in the same category as Anthropic’s computer-use capabilities that debuted with Claude in late 2025.

On benchmarks, the gap has narrowed significantly. Codex running on GPT-5.5 scores 88.7% on SWE-bench Verified, slightly ahead of Claude Code on Opus 4.6 at 87.6%, according to ShareUHack’s analysis. Both scores represent substantial gains from early 2025 baselines.

Integration breadth also differs. Codex connects to VS Code, JetBrains, GitHub, Slack, Linear, and any MCP-compatible tool. Claude Code adds Chrome, GitLab, and automated PR review natively. Both support parallel agent execution across multiple files, though they brand it differently (Codex uses parallel agents; Claude Code uses Routines).

Pricing Splits the Market

The most significant competitive lever may be access. Codex is included in every ChatGPT plan, including the free tier, with an active 2x usage promotion running through May 31, 2026. Claude Code requires a paid Anthropic plan. For individual developers and indie makers, that pricing gap matters more than marginal benchmark differences.

Anthropic countered access constraints with infrastructure: the SpaceX Colossus deal doubled Claude Code rate limits for paid users. But Claude Code’s Max plan users have reported hitting rate limits faster in recent weeks, according to Softr.

The Competitive Dynamic

Anthropic’s Claude Code grew 80x in its first year, dominating the developer-agent category since early 2025. OpenAI’s response has been systematic: feature parity first, then price competition through free-tier access. The question is whether OpenAI can convert ChatGPT’s existing 300+ million user base into Codex adopters, or whether Claude Code’s head start in developer trust and workflow integration creates a durable moat.

For teams evaluating both, the practical advice from independent reviewers is converging: use Codex for backend-heavy, precision-critical work; use Claude Code for full-stack projects where intent inference saves prompting time. Running both is viable but requires comfort with debugging generated outputs and managing infrastructure decisions independently.