.png)
Claude Code has pulled ahead of Codex in global usage according to the new CostHawk leaderboard first posted on Hacker News. The site aggregates real consumption data across API calls and shows Claude Code holding a 57% share versus Codex at 43% for the latest 30-day window.
Tool: CostHawk Leaderboard | Models tracked: Claude Code, Codex | Data window: 30 days | Source: costhawk.ai/leaderboard
What the Leaderboard Shows
The dashboard breaks usage into daily active keys, total tokens processed, and average request size. Claude Code leads in token volume at 2.8 trillion tokens over the tracked period while Codex sits at 2.1 trillion. Average request length for Claude Code is 1,240 tokens compared with 980 for Codex.
Regional splits reveal Claude Code stronger in Europe and North America while Codex maintains higher share in Asia-Pacific markets.
Key Numbers and Trends
| Metric | Claude Code | Codex |
|---|---|---|
| Share of calls | 57% | 43% |
| Tokens processed | 2.8T | 2.1T |
| Avg request size | 1,240 tokens | 980 tokens |
| Peak daily keys | 184k | 141k |
Early HN comments noted the gap widened after Claude 3.5 Sonnet updates in late 2024.
How to Check the Data Yourself
Visit costhawk.ai/leaderboard and select the 7-day or 30-day toggle. Export buttons provide CSV files with daily breakdowns. No login is required for basic views; paid tiers unlock per-key filtering.
Pros and Cons
- Real consumption numbers instead of self-reported benchmarks
- Daily updates with 30-day rolling window
- Free CSV export for analysis
- Limited to two models only
- No public methodology on data sourcing
- No historical data before September 2024
Alternatives and Comparisons
LMSYS Chatbot Arena focuses on preference votes rather than actual API spend. Artificial Analysis tracks latency and price but lacks usage volume. The CostHawk approach sits between the two by showing what developers actually call at scale.
| Feature | CostHawk | LMSYS Arena | Artificial Analysis |
|---|---|---|---|
| Usage volume | Yes | No | No |
| Latency metrics | No | Partial | Yes |
| Free export | Yes | No | Yes |
| Update frequency | Daily | Daily | Weekly |
Who Should Use This
Teams evaluating API spend should check the leaderboard weekly to spot adoption shifts. Researchers tracking model preference across regions gain concrete numbers without running their own surveys. Individual developers deciding between providers can skip it and test both models directly on their workloads.
Verdict
The CostHawk leaderboard supplies the first public view of real Claude Code versus Codex consumption at global scale. It fills a gap between preference arenas and price sheets by showing which model developers actually keep calling.
Top comments (0)