GPT-6 Astra edges past Claude Fable 5.1 on WebDev Arena

As of September 5, OpenAI's GPT-6 Astra leads Code Arena's WebDev leaderboard with 1,797 points, 35 ahead of Anthropic's Claude Fable 5.1 at 1,762. The result was published on Threads, which credits the OpenAI model with the top position on a board rated by crowdsourced votes.

At a glance

• The WebDev board ranks models on a crowdsourced Elo-style rating, and the current order rests on more than 650,000 votes cast across a field of 126 models on Code Arena.
• Alongside the ranking, the post describes Astra as reshaping the Pareto frontier as the best-performing model at $40 per million tokens, a rate it says matches the latest Claude model's pricing.
• Claude Fable 5.1 arrived days earlier and took first place in the Artificial Analysis Intelligence Index, leaving the two models at the top of two separate public rankings at once.

For teams picking a model for front-end work, the matching $40 per million token rate takes price out of the comparison and leaves a 35-point gap on a preference-driven board as the deciding signal, which is a thin basis for rewriting a stack. The wider picture reads as unsettled: one model leads WebDev, the other tops the Artificial Analysis index. A 35-point spread on a preference-built rating is a narrower signal than it looks at the top of a 126-model field.

GPT-6 Astra holds 1,797 points to 1,762 for Claude Fable 5.1

GPT-6 Astra sits at 1,797 points at the top of Code Arena's WebDev leaderboard, with Claude Fable 5.1 directly behind it at 1,762. The distance between the two entries is 35 points on the board's scale, the margin that puts the OpenAI model in first place in the September 5 reading of the board.

The ordering comes from a crowdsourced Elo-style rating system rather than a fixed benchmark suite, with positions set by user votes comparing model outputs. The board carries more than 650,000 votes spread across 126 models, and the two leaders sit above every other entry in that field on the current reading.

Claude Fable 5.1 shipped days before that reading and took first place in the Artificial Analysis Intelligence Index, a ranking maintained separately from Code Arena's WebDev board. On the WebDev board itself, the Anthropic model holds second position, with its 1,762 points recorded on the same crowdsourced scale.

The post calls Astra the best-performing model at $40 per million tokens

The result is framed in cost terms as well as ranking terms. The post describes GPT-6 Astra as reshaping the Pareto frontier and as the best-performing model at $40 per million tokens, the rate it says the latest Claude model also carries.

It also reshapes the Pareto frontier as the best-performing model at $40/Mtoken, which matches the latest Claude model pricing.

A Pareto frontier in this context is the set of models that no other model beats on both quality and price at once, and a model reshapes it by outscoring everything available at or below its own rate.

The Pareto frontier claim is made for GPT-6 Astra alone, with the Claude model named only as the price reference. At that shared quoted rate, the separation the post reports between the two is the 35-point margin on the WebDev leaderboard rather than any difference in cost per token.

Whether the board reaches 2,000 The post closes on the question of whether ratings hit 2,000 points before the end of the year, with the leader now at 1,797 and the runner-up at 1,762. Crowdsourced Elo scores move as new votes land and as new models join the pool, so the current order describes the September 5 reading rather than a settled standing.