GPT-6 Astra takes WebDev Arena #1, 35 points over Fable 5.1

GPT-6 Astra (Max) landed at 1,797 on Code Arena's WebDev leaderboard, 35 points clear of Claude Fable 5.1 (Max) at 1,762 and 109 clear of Claude Opus 5 at 1,688. It's the highest mark any model has posted there — five days after Astra shipped on 3 September.

The pricing detail is the part worth staring at. Both Max tiers charge $10 per million input tokens and $50 per million output, and both carry 1M-token context windows. OpenAI didn't buy the lead with a pricier model; it took the quality crown at the price Anthropic had already set. And on a board scored by 650,000+ human votes across 126 models, that's a preference win on messy, iterative frontend work — not a fixed-dataset benchmark.

Whether the board hits 2,000 by December is a different question. Arena Elo isn't a capped scale — it drifts upward as stronger models join the pool — so a 200-point climb in four months is plausible arithmetic and a bad proxy for 200 points of real capability. The number that means something is the gap between #1 and #2, and 35 Elo is roughly a 55/45 head-to-head preference. That's close.

Why it matters: when the top two models cost exactly the same, the leaderboard stops being a shopping list and starts being a coin flip you re-run every month.