Skip to content
IIAIT

AI Today · 28 October 2026

OpenAI releases a cheaper computer-use model and Cloudflare discloses a prompt leak in Workers AI

Today: a lower-priced OpenAI model for everyday computer work, a Cloudflare bug that showed some customers fragments of other customers' requests, Opera running AI models locally in its browser, and an expanded Claude offering for nonprofits.

  1. 01

    OpenAI releases GPT-6.1 Terra for computer use at half the API price of its predecessor

    OpenAI released GPT-6.1 Terra on 26 October, a mid-tier model aimed at operating apps on screen, filling in forms, coding and other routine work. It is rolling out to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and through the API as gpt-6.1-terra at 1 dollar per million input tokens and 5 dollars per million output tokens, which OpenAI says is 50 percent below GPT-5.6 Terra and GPT-6.1 Sol. OpenAI claims it matches GPT-6 Astra on its agent tests at a tenth of the per-token price; these are the company's own benchmarks. OpenAI also classifies the model as reaching the Critical cybersecurity level of its Preparedness Framework and says it keeps the safeguards built for GPT-6 Astra.

    What it means for you

    If you pay for automated or agent-style work through the API, Terra may cut costs for routine tasks such as forms and document handling; test it on your own workflows before switching, since the published results come from OpenAI. For the hardest coding or research work, OpenAI itself still points to its larger models.

    Sources

  2. 02

    Cloudflare: a bug in its hosted GLM-5.3 model returned other customers' data in 155 responses

    Cloudflare disclosed on 26 October that between 27 August and 9 September, its Workers AI deployment of the GLM-5.3 model sometimes returned data from another customer's request made just before. All 155 affected responses followed prompts longer than 681,280 tokens. The company traces it to a configuration error, where the model accepted longer contexts than its cache could hold, combined with a bug in the open-source SGLang inference engine. A security researcher reported the problem on 8 September and the model was fixed and restored on 9 September. Cloudflare says no other Workers AI model was affected and that it found the pattern consistent with normal long-context use rather than deliberate exploitation.

    What it means for you

    When you send data to a hosted AI model, keeping it separate from other customers depends on the provider's engineering, not only on its contract terms. Avoid sending more sensitive data than a task needs, and ask your AI providers how they isolate customers and how they would notify you of an incident.

    Sources

  3. 03

    Opera ships local AI models in its browser as an opt-in preview

    Opera announced on 27 October an early preview of WebAI in Opera One, which lets websites and extensions use an AI model that runs on your own computer through the browser. It is off by default; nothing is downloaded until you enable it in settings, and you can choose which sites may use it. At launch it offers Google's Gemma and, on Macs, Apple's on-device model, on 64-bit Windows PCs with Intel or AMD processors and on Apple silicon Macs. Opera says data handled by the local model stays on the device, and Opera GX and Opera Neon are due to follow.

    What it means for you

    On-device AI can keep text such as drafts or internal documents off external servers, which can help with data protection. Treat it as a preview: check your organisation's browser policy before enabling it, and grant site permissions only to sites you trust.

    Sources

  4. 04

    Anthropic expands Claude for Nonprofits with a dedicated plugin and a lower team-plan minimum

    Anthropic announced on 26 October a Claude for Nonprofits plugin with skills and connectors for tasks such as fundraising, board work and hiring, installed at no extra cost for verified nonprofits on Team and Enterprise plans. Nonprofit plans now include the same models as standard Team and Enterprise plans, with Claude Code included, and the Team plan is open to organisations with as few as two people. Existing discounts remain: up to 75 percent, and about 90 percent for verified nonprofits in low- and middle-income countries. Anthropic is also running two webinars for administrators and staff on 10 and 12 November.

    What it means for you

    If you work for a small nonprofit, a discounted team plan may now be within reach, but agree a short internal AI policy first on what data may be shared, especially donor and beneficiary data covered by GDPR.

    Sources