Breaking

GPT-6 Astra Tops Image-to-WebDev Arena as Claude Fable 5.1 Takes Second

September 15, 2026 at 17:21 EDT

Foundation Models
Software Dev & Coding
AI Agents

GPT-6 Astra (Max) has moved into first place on Arena's Image-to-WebDev ranking with 1,733 points, putting OpenAI's newest model ahead of Claude Fable 5.1 (Max) at 1,710. The update adds another public signal that frontier models are being judged not only on chat and coding tests, but on their ability to turn visual designs into working web interfaces.

September 15, 2026 · Arena benchmark

GPT-6 Astra takes the image-to-web lead

OpenAI's GPT-6 Astra (Max) reached 1,733 points, narrowly ahead of Claude Fable 5.1 (Max) and well above GPT-5.6 Sol (xHigh).

1,733

GPT-6 Astra score

1,710

Claude Fable 5.1 score

Point lead over No. 2

129

Point lead over GPT-5.6 Sol

Leaderboard scores, drawn to scale

Astra's lead is narrow over Claude Fable 5.1, but wider against OpenAI's prior GPT-5.6 Sol entry.

Why builders care

The task tests whether a model can read visual hierarchy, spacing and interface structure, then turn that understanding into editable front-end code.

What still limits deployment

Cost, enterprise access controls, safety interventions and long-running agent reliability remain key constraints for production workflows.

The benchmark is a snapshot, not a finish line

Signal

Image-to-web ability is becoming a visible frontier-model battleground.

→

Decision

Teams will weigh score gains against price, availability and operational risk.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.