AI NewsOctober 10, 2026

OpenAI’s Decisions API, in public beta since October 6, returns a yes/no probability, a pick, or a score about text or images. It runs only on GPT-6 Luna.
Ask most AI a question and it writes a paragraph. OpenAI’s Decisions API only returns a probability, a choice, or a score about text or an image, up to 10x faster than its standard API. Think ticking a box, not writing an essay.
It runs on GPT-6 Luna for $0.10 per million input tokens (chunks of text), with no charge for output. Sorting tickets or screening images becomes cheap enough to run on everything.
Key Capabilities:

• Three answer types: A yes/no probability, a pick from a list, or a numeric score.
• Text or images: Accepts either or both in one request, so photos can be sorted too.
• Still beta: OpenAI expects general availability in the coming weeks, and only GPT-6 Luna works for now.

Microsoft-Decision-1 picks from a fixed set of options and scores each with a probability instead of writing text. It is built on a 9B open model and is live in Microsoft Foundry.
Most AI answers by writing. Microsoft-Decision-1 skips that and picks from a fixed set of choices, giving each a probability, like a multiple-choice grader. Microsoft says it runs 35x faster than GPT-6 Sol.
That matters because much of business AI is sorting, not writing: routing tickets, flagging faults, checking quality. Microsoft already tests it internally on incident response and quality control.
Key Capabilities:

• Pick, don’t write: Returns a probability for each option, so software can act on the answer instantly.
• Small base: Built on the open Qwen3.5-9B model, retrained by Microsoft for decision scoring.
• Cheap and live: Available now in Foundry at $0.042 per million input tokens (chunks of text), output free.

Alibaba’s Qwen team released Turbo, a faster version of its 7B image model. It makes 2K images from text in 8 steps and edits them from plain-language instructions.
AI image models build a picture by cleaning up static in rounds, like a painter adding coats. Alibaba’s new Qwen-Image-2.1-Turbo needs only 8 rounds, yet still makes 2K images from text and edits them from plain instructions.
The model weights are open for download, and hosted Pro and Turbo APIs are live. Fewer rounds means less computing per image, which makes image tools cheaper to run for small teams.
Key Capabilities:

• 8 rounds only: Creates and edits images in 8 passes on the same 7B-size model as the original.
• Edit by typing: Continues a picture through plain-language requests, from adding accessories to changing the whole scene.
• Two ways in: Download weights from Hugging Face or ModelScope, or call the hosted Pro or Turbo API.