GPT‑6 Astra

OpenAI has launched GPT‑6 Astra, initially available to a limited set of organizations and, in the coming days, to all ChatGPT Plus, Pro, Business, and Enterprise subscribers, as well as through its API and AWS infrastructure. The model is priced the same as Claude Fable 5 and 5.1: $10 per million input tokens and $50 per million output tokens, positioning it as a direct competitor to the Fable family.

In internal tests, Astra scores 99.9 % on the ARC‑AGI 3 benchmark, though that result was achieved using an in‑house “Provider Adapter harness” that preserves reasoning state across requests and compresses long conversations. In security tasks, the model outperforms its rivals, earning perfect scores on ExploitBench and far superior results to GPT‑5.6 Sol on other vulnerability and binary‑analysis metrics.

The new model also shows significant gains in handling long contexts, reaching 100 % on tests with 256,000 to 512,000 tokens and maintaining performance above 96 % with up to one million tokens. However, in the intelligence index compiled by Artificial Analysis, Astra trails behind Claude Fable 5.1 and Meta’s recent Muse Spark 1.3. In assisted programming, Astra stands out for cost efficiency, surpassing Fable in price‑to‑score ratio.

This news matters because a model with stronger long‑term reasoning and greater robustness in security tasks could influence the adoption of generative AI in critical enterprise environments, driving new applications and raising the bar for industry competition.

Read the original article on Simon Willison

This summary is an informational synthesis produced by dataqbs.com. All rights to the original content belong to its author and the cited media outlet. We act solely as curators of technology news and claim no authorship.