Claude Sonnet 5.5 takes third place on Code Arenaās WebDev leaderboard
Anthropic's mid-tier model scored 1786 points on Arena.ai's community-voted coding rankings, close behind two pricier flagship rivals
Anthropic’s Claude Sonnet 5.5 has landed in third place on Arena.ai’s Code Arena WebDev leaderboard. Running with xHigh reasoning, it scored 1786 points.
That would be a respectable result for any model. It stands out more for a mid-tier option sold at half the price of its bigger sibling, sitting a few points behind the flagships.
How the leaderboard shakes out
The top spot belongs to Claude Opus 5.5 Max, which scored approximately 1818 to 1820 points. In second is OpenAI’s GPT-6 Astra Max, at around 1792 points.
Sonnet 5.5 trails the OpenAI model by roughly six points. At the very top of a crowded table, that is a narrow margin.
The WebDev arena focuses on front-end web development. Models tackle tasks that demand complex, multi-step reasoning and the use of tools, mostly built with HTML and React.
The work runs as multi-turn, agentic workflows. Models plan, build and revise across several steps instead of answering one prompt and calling it a day.
Arena.ai builds the rankings from community votes. Users compare outputs from two models side by side and pick the one they prefer.
As of late September 2026, the leaderboard had collected more than 800,000 votes across over 100 models.
A cheaper model punching above its price tag
Sonnet 5.5 launched around September 28, 2026. It arrived as Anthropic’s mid-tier offering, the option between lightweight models and the top-shelf Opus line.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
Pricing is set at $2 per million input tokens and $10 per million output tokens. That works out to half the cost of Opus 5.5.
The leaderboard result is not the only data point working in Sonnet’s favor. The model scored 70.6% on Terminal-Bench 4.0, a test of agentic coding performance.
On that same benchmark, Opus 5.5 scored 66.4%. The cheaper model came out ahead by 4.2 percentage points.
The two results measure different things. Terminal-Bench is a fixed benchmark, while Code Arena reflects human preference on practical web tasks.
On human preference, Opus 5.5 Max still leads. On the terminal-based agentic test, Sonnet 5.5 took the edge.
Why community voting matters here
Arena.ai’s approach is built around practical coding tasks instead of static benchmarks. The ranking reflects which outputs real users actually preferred.
Pairwise voting asks a simpler question: which of these two results is better? With hundreds of thousands of those judgments, the rankings capture something closer to everyday usefulness.
What this means for developers and the AI coding market
For developers, the math is fairly simple. A model that sits within single digits of second place, at half the price of Anthropic’s flagship, becomes a strong default for many coding workloads.
It also complicates Anthropic’s own product lineup. When Sonnet outscores Opus on Terminal-Bench 4.0, customers will reasonably ask what the extra spend on Opus buys them.
The likely answer, based on the Code Arena rankings, is the top spot on human-judged web development. Opus 5.5 Max holds a lead of more than 30 points over Sonnet 5.5 there, a wider gap than the one separating Sonnet from second place.
For OpenAI, GPT-6 Astra Max holds second place, but it sits between two Anthropic models on a leaderboard that developers watch closely.