GPT-6 Astra is elite at agents, math and cyber — but general intelligence stays flat and the price is 2.5×

Asked:

how good is the astra by chatgpt

An early read on OpenAI's GPT-6 Astra, released 3 September 2026, across six evaluation categories — coding agents, general intelligence, reliability, advanced math, cybersecurity, and cost — mixing OpenAI-reported benchmarks with independent Artificial Analysis testing and published API pricing.

Independent (Artificial Analysis) OpenAI-reported Pricing (lower is better) Predecessor GPT-5.6 Sol

Independent testing puts Astra's Intelligence Index at 61 — exactly level with GPT-5.6 Sol, no generational leap. artificialanalysis.ai

Hallucinations drop from 92% to 51% on AA-Omniscience — a major reliability gain, yet still far from error-free. artificialanalysis.ai

OpenAI reports 100% on ExploitBench (vs 78.5% for Sol) and classifies Astra at Critical cyber capability, with unusually strict safeguards. openai.com

Standard pricing is $10/M input and $50/M output tokens — 2.5× GPT-5.6 Sol — rising to $20/$75 for long context. developers.openai.com

The six categories

Category	Key metric	Verdict	Comparison	Source

Who should use it

Reach for Astra when the task is genuinely hard and agentic: multi-step coding agents, difficult computer-use workflows, frontier math, and security research — roughly an 8/10 there. For everyday chat and cost-sensitive work it is closer to 6.5–7/10: independent intelligence scores match its cheaper predecessor, so the 2.5× price is hard to justify outside its specialist strengths.

Six evaluation categories, one row each, assessed 2026-09-04, one day after Astra's 2026-09-03 launch. Metrics mix OpenAI-reported benchmarks (FrontierMath Tier 4, ExploitBench) with independent Artificial Analysis indices and hallucination testing, plus published API pricing; each bar shows the headline number in its metric's own unit, so bars compare within a category, not across. Early assessment — treat OpenAI-reported results with more caution than independent ones.

This report was generated automatically by Keenable SELECT at a user's request, from publicly available web sources linked herein. Keenable does not review, verify, or endorse its contents and makes no representation as to accuracy, completeness, or timeliness; AI-based extraction may contain errors. Nothing in this report is investment, legal, financial, or other professional advice. All trademarks and referenced content remain the property of their respective owners; no affiliation or endorsement is implied. To report an error, rights concern, or request removal: legal@keenable.ai.