$80,000. That's what some individual Meta engineers have spent on Claude API compute - and the company is encouraging it.
Meta has given its engineering teams unlimited access to Claude tokens and, in a move that says a lot about how Big Tech views AI costs right now, created an internal leaderboard tracking who racks up the most spend. It's not a warning system. It's a scoreboard.
The approach signals a specific bet: that the productivity gains from heavy AI usage outweigh even six-figure per-person compute bills. Early reports from Meta engineers suggest the math is more complicated than the hype suggests. AI-assisted coding roughly doubles output for experienced developers, but that 2x comes with real tradeoffs - time lost to naive solutions the model generates, wheels it unnecessarily reinvents, and code that works but misses existing internal patterns.
What $80K in Tokens Actually Buys
At Anthropic's current API pricing for Opus 4.6 ($15 per million input tokens, $75 per million output tokens), $80,000 buys roughly 1-5 million back-and-forth exchanges depending on conversation length and complexity. That's an engineer essentially pair-programming with Claude for every waking work hour across several months.
The more telling number isn't the ceiling but the average. Most Meta engineers using Claude aren't anywhere near $80K. The leaderboard extremes likely represent developers on massive codebases using Opus for everything, including tasks where cheaper models would suffice.
For the rest of the industry, Meta's experiment matters because it provides real data on a question every company is asking: what's the right budget for AI developer tools? The answer appears to be "more than you think, but with diminishing returns." A developer spending $5,000/month on Claude is probably getting strong ROI. Whether the jump from $5K to $80K delivers proportional value is the open question Meta's leaderboard might eventually answer.