Sunday, 6 September 2026

OpenAI opened GPT-6 Astra to developers on 5 September 2026, two days after unveiling the model, and it is now callable through the API as gpt-6-astra [1]. According to [1], standard pricing is $10 per million input tokens and $50 per million output tokens, with cached input reads dropping to $1. The model carries a 1,050,000-token context window and can return up to 128,000 tokens in a single response [1][2].

The output price is the headline. At $50 per million tokens Astra sits well above the models beneath it, so the caching discount is what keeps it affordable [2]. According to [1], a workload that reuses a large system prompt or a fixed retrieval context is billed mostly at the $1 cached input rate rather than $10, and that is where the saving comes from. Reviewers who tested the endpoint note the API now returns only a summary of the model's reasoning, not the raw steps [1].

For people who buy AI through other software rather than calling the API directly, the read is about where prices settle. A more capable model at a higher list price, discounted heavily for cached context, signals that the next round of features you use may cost more per call unless the vendor leans on caching. Whether that reaches your bill depends on how the tool you pay for is built [2].

The Seattle Times and Newsday sued OpenAI and Microsoft on Friday, 4 September 2026, in the U.S. District Court for the Southern District of New York, accusing the companies of using their journalism to train AI systems without permission [1]. According to the suit, the two companies scraped the newspapers' articles, including material behind paywalls [2], and fed them into the datasets behind ChatGPT and Microsoft Copilot [1].

The complaint describes generative AI as "a snake eating its own tail" that hands back "copies and derivative imitations" of human-written work and could "destroy the very organizations" that produce it [1]. The papers argue the AI products harm the advertising revenue and business model that fund their newsrooms [2]. Microsoft said it was surprised by the suit but open to discussing solutions, according to [1]; the two companies have previously funded some Seattle Times journalism projects [1].

The filing widens a fight that opened when The New York Times sued the same two companies in 2023 [1]. According to [2], this adds two more newsrooms to a growing line of publishers taking AI firms to court. The underlying question — whether training on copyrighted work without a licence is fair use — is still unsettled, and each new suit raises the odds it is decided in court rather than by a deal.

Three hikers were rescued on Mount Shasta in northern California after a climb that ran badly over schedule, and the Siskiyou County sheriff's office pointed partly at AI trip planning [1]. According to [1], the group set out around 3 a.m., did not reach the summit until about 7 p.m. — hours past the recommended midday turnaround — and spent the night in a canyon before rangers and volunteers brought them out the next morning.

The sheriff's office said the hikers "were advised by Gemini to bring far less food and water than their group required," and that the shortfall mattered once an ascent planned for hours became a multi-day ordeal [1]. Its advice was blunt: call the local Forest Service ranger station before a trip, and "never rely solely on AI for your trip planning" [1]. The incident was first reported by the Chicago Tribune [2].

The lesson is not that AI planning tools are useless but that they answer with a confidence they have not earned: a model does not know the mountain, the season or your party, and it is not the body that comes to find you. For anything where being wrong is dangerous, the check belongs with the authority that owns the ground.