GPT-6 Astra raises the bar on Image-to-WebDev benchmark

OpenAI's GPT-6 Astra (Max) has taken the top spot on the Image-to-WebDev Arena leaderboard with 1,733 points, edging out GPT-5.6 Sol (xHigh) by 129 points and the runner-up by just 23. The tight clustering at the summit signals that the race among the major labs has moved from generational leaps to fine-grained optimisation.

Anthropic's Claude Pavell 5.1 (Max) sits second with 1,710 points, an 87-point gain over Pavell 5 and 45 points clear of Claude Opus 5 (Max) in third at 1,665.

Meta AI's Muse Spark 1.3 (Max) debuts directly in fourth with 1,645 points, while Z.ai's GLM-5.3-Flash cracks the top ten at 1,588. The notable detail: three of the four new entrants — GPT-6 Astra, Muse Spark and GLM-5.3-Flash — occupy the Pareto frontier of performance versus cost, meaning no other model beats them on both metrics simultaneously.

The pricing data reveals order-of-magnitude gaps. GPT-6 Astra charges $40 per million tokens for the highest score. Muse Spark delivers 1,645 points at $3.5 per million, a roughly 91% discount for a roughly 5% performance concession. GLM-5.3-Flash disrupts the market at $0.21 per million tokens (under one shekel), delivering 1,588 points — about 92% of the leading score at one-tenth of one percent of the price.

Image-to-WebDev Arena does not test general knowledge or creative writing. Its methodology ranks models on the ability to turn screenshots and images into functioning websites, including agentic coding workflows that require multi-step reasoning and tool use. It is a practical test of automated front-end engineering, not a chat benchmark, so the results speak directly to developers building design-to-code tooling.

For teams that need maximum performance and will pay a premium, GPT-6 Astra is the only option at the peak. For aggressive cost-performance, Muse Spark offers 95% of the capability at 9% of the cost. For massive workloads on a minimal budget, GLM-5.3-Flash delivers a usable result at near-zero cost. All three sit on the Pareto frontier, so the choice comes down to where you land on the budget curve.