Subscribe
AI
2 min read

Claude Fable 5.2 Tops the Code Arena WebDev Leaderboard After a 20 Percent Price Cut

Anthropic's Claude Fable 5.2 landed at number one on the Code Arena WebDev leaderboard with a rating of 1,832, according to Arena's announcement, extending a lead the company built in September when Fable 5.1 topped the same board. The release that put it there came ten days ago, and the leaderboard is the first independent scoreboard since. The numbers At its November 9 launch, Fable 5.2 arrived with a 20 percent price cut: $8 per million input tokens and $40 per million output, down from $1

Claude Fable 5.2 Tops the Code Arena WebDev Leaderboard After a 20 Percent Price Cut

Anthropic's Claude Fable 5.2 landed at number one on the Code Arena WebDev leaderboard with a rating of 1,832, according to Arena's announcement, extending a lead the company built in September when Fable 5.1 topped the same board. The release that put it there came ten days ago, and the leaderboard is the first independent scoreboard since.

The numbers

At its November 9 launch, Fable 5.2 arrived with a 20 percent price cut: $8 per million input tokens and $40 per million output, down from $10/$50 on Fable 5.1, per Anthropic's release notes coverage. Anthropic also claims outputs up to 30 percent faster than the 5.1 models, plus an optional fast mode at $16/$80 that raises generation speed up to 2.5x. The model keeps a 1-million-token context window and runs on the Claude API, Claude apps, Claude Code, and the three major clouds. Datanorth's breakdown notes Anthropic's internal figures put it at 69.9 percent on FrontierSWE v2, ahead of Claude Opus 5.5 at 62.3 percent.

On the live WebDev leaderboard, the xHigh effort setting of Fable 5.2 sits in front of a stack of former leaders:

  • Claude Opus 5.5 (Max) at 1,818
  • Gemini 4 Carbon (High) at 1,768
  • GPT-6 Astra (Max) at 1,767

The gap to second place is double digits, but the field is closing: Arena's board has passed through several number-one models in 2026, including Moonshot's open-weights Kimi K3, which briefly topped the Frontend Code Arena in the summer.

Olivetti-style illustration of a podium ranking board with model tiers

Why the cut matters more than the crown

The price move is the durable fact here. Fable 5.2 now lists below GPT-6 Astra while holding the top coding rating, and Anthropic says it consumes fewer tokens than its predecessor on hard tasks at medium effort, which compounds the sticker discount into a real per-task saving. On the model overview, Anthropic positions Fable as its line for demanding reasoning and long-horizon agentic work, with the sibling Mythos 5.2, the same model without cyber and biology restrictions, staying inside its vetted verification programs.

One caveat on the leaderboard itself: Arena ratings measure human preference in head-to-head web development tasks, not task completion or cost. The board has moved fast this year, and a 1,832 today is a snapshot, not a property of the model. Still, shipping a price cut and a leaderboard top within one IPO week is a coherent message about where Anthropic thinks the competition is: sustained agentic coding, at a billable rate.

Sources

AILLMEvaluations