Breaking

OpenAI Launches GPT-6 Astra With Computer-Use Gains and a Mixed Benchmark Record

September 3, 2026 at 16:25 EDT

Foundation Models
AI Agents
Security

OpenAI has released GPT-6 Astra, a model built to operate computers and carry out long, multistep professional work. The company presents it as its most intelligent and aligned model, but published comparisons show strong category wins alongside clear gaps against Anthropic's Claude Fable 5.1 and Claude Opus 5.

September 3, 2026 · OpenAI

GPT-6 Astra enters a mixed benchmark field

OpenAI says Astra is its most intelligent and aligned model, built for computer use and long-horizon work. Its headline gains are substantial, but rival models still lead on several composite and coding measures.

Where the leaderboard is less clear

Artificial Analysis Intelligence Index v4.1.1 · higher is better

Astra
61.2

Claude Fable 5.1
65.7

Each block represents roughly five index points.

FrontierMath Tier 4

97.6%

Academic evaluation

ARC-AGI-3

99.9%

Abstract reasoning

Output price

$50

Per 1M output tokens

What looks strong

Computer use across forms, CRMs, calendars, research, and quality assurance
Customer workflows report 50% fewer manual fixes and 41 documents reviewed in minutes

What remains difficult

Several composite and coding comparisons still favor rival models
Critical cyber capabilities require restricted rollout and stronger monitoring
Safety checks can pause legitimate work while reasoning is harder to monitor

Continue reading

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