$1.87 million. That's how much 572 developers collectively spent on AI coding tools, according to usage data tracked by tokscale, an anonymous logging tool that reads local usage data from AI coding assistants without capturing API keys or actual code.
The numbers paint a split picture of the AI coding market: OpenAI handles 56.8% of all tokens (1.89 billion) but only 42.4% of total spending ($677,700). Anthropic processes just 34.8% of tokens (1.16 billion) but takes 54.1% of the money ($864,600). In plain terms, developers send more work to OpenAI, but they pay significantly more per unit of work when using Anthropic's models.
The Model-Level Breakdown
At the individual model level, Claude Opus 4.6 alone accounts for $303,200 in spending, while GPT-5.4 leads OpenAI's side at $381,600. The next tier down shows Claude Opus 4.5 at $131,700 and GPT-5.3-codex at $213,000. Google barely registers at 2.44% of tokens and 1.5% of spend. xAI is a rounding error at 0.36% token share.
The average developer in this dataset burns through 5.8 billion tokens and spends $3,300 total, with median daily usage landing at 4,200 messages and 372.7 million tokens. That median daily figure suggests these are heavy AI-assisted coding users, not casual ChatGPT prompters.
Which Tools Developers Actually Use
Codex CLI dominates with 50.7% market share across 393 users. Claude Code sits at 21.6% (502 users - more individual users but less total volume per person). OpenCode takes 19.5% with 390 users, while Cursor - probably the most well-known AI coding tool among general audiences - captures just 2.8% share from 55 users.
That Cursor number deserves context. This dataset skews toward CLI-first, terminal-native developers who opt into anonymous tracking. Cursor's actual user base is much larger, but this crowd prefers tools that live in the terminal rather than an IDE.
What This Actually Tells Us
The sample size of 572 developers is small and self-selecting. These are power users who installed a token-tracking tool, so the spending figures represent the ceiling, not the average. But the relative market share data is still useful: it shows that among the most committed AI coding practitioners, the race is between OpenAI and Anthropic, with Google a distant third.
The cost-per-message average of $0.05 is also telling. At 4,200 messages per day, that's roughly $210 daily or $4,400 monthly for heavy users. For a senior developer whose time costs $80-150/hour, that math works out if AI assistance saves even 30-60 minutes per day. The extrapolated monthly run rate across all tracked users: $3.7 million.