AI Release Roundup Fall 2026: Every Major Model, Chip and Tool in One Place
The fall of 2026 has been the most intense period of AI releases yet. One tracker verified 108 AI tool and model launche...
A new category of AI model quietly became one of the biggest trends of fall 2026. Instead of writing paragraphs, decision models answer bounded questions with probabilities: yes or no, pick one of several options, or give a score. In the space of a few weeks, OpenAI, Google, Perplexity, Liquid AI, HeyGen, Sarvam, Kakao and Nace AI all released decision models or APIs. If you build automations or manage high-volume workflows, this category may save you more money than any new flagship model.
A traditional language model generates text one token at a time. If you ask it whether an email is a complaint, it writes an answer, which you then have to parse. A decision model takes a context, such as a document, an image or JSON data, plus one or more typed questions, and returns a calibrated probability for every possible answer in a single pass.
That design has three big advantages:
Typical uses include routing support tickets, checking documents against policies, grading responses, moderating content, triaging intake forms and judging whether an agent completed a task correctly.
OpenAI made its Decisions API generally available in October on gpt-6-luna. A request returns a probability, a choice or a score for bounded questions over text, images or JSON, up to ten times faster than the Responses API. General availability added prompt caching, integration with GPT-Live, US and EU inference residency and up to 400 questions per request.
On November 11, Google launched the Gemini Decisions API powered by Gemini Decision 1. One multimodal input of up to 256,000 tokens, including text, images, audio, video or PDFs, can be checked against up to 1,000 independent questions in a single pass. Each answer includes probabilities and a confidence score. Google quotes 150 to 250 milliseconds of latency at 0.10 dollars per million input tokens, with zero output tokens billed.
Perplexity released Decider v1.3 and Decider Flash v1.3 on November 12, at 27 billion and 4 billion parameters respectively. They add a locate capability, which extracts verbatim text spans by type or instruction, alongside yes or no, multiple-choice and rubric decisions. Weights are available under Apache 2.0, and the models are offered through Perplexity's Decisions API and OpenRouter. Perplexity also uses Decider inside its new Monitor API, covered in our article on Perplexity Teammates and Monitor.
Liquid AI released d1.2 and d1.2-ultra on November 16, adding audio input and five new answer types: reading a value, finding every passage of a kind, locating objects with bounding boxes, estimating a number with a range and selecting every option that applies. Liquid reports that d1.2-ultra ranks first among decision models on the Artificial Analysis Decision Index. Pricing is input-only at 0.02 and 0.042 dollars per million tokens.
Decision models can screen incoming documents for privilege, classify matters by practice area or check whether contracts contain required clauses. A typical setup asks a dozen yes or no questions about each document, then routes only flagged items to a lawyer. Pair this with workflows like the case summary auto-builder and document review resources.
Use decision models to flag unusual transactions, verify that invoices meet policy or classify expenses. The IFRS and GAAP compliance checker agent shows the type of rule-based checking that benefits from fast, structured answers. Browse more in internal audit resources.
Score content against brand guidelines, moderate comments or classify leads by intent. See social media tools for related workflows.
One of the most valuable uses is judging the output of other AI systems. A decision model can check whether an agent's draft answers the question, follows the format and avoids prohibited content, all before a human sees it.
Automation platforms make it easy to drop a decision step into existing workflows. Explore our n8n workflow templates and AI agent collection for patterns.
Usually, yes. Many charge only for input tokens and return results faster, which reduces both cost and latency.
No. They complement them. Use decision models for classification and routing, and generative models for drafting and analysis.
It depends on your data and environment. Google and OpenAI offer large hosted options, while Perplexity, Sarvam and Nace offer open weights you can host yourself.
Decision models turn many AI tasks from expensive text generation into fast, cheap and reliable structured answers. For professionals running high-volume workflows, they are one of the most practical AI developments of the year. Start with one routing or screening task, measure it carefully against human decisions and expand from there.
Elevate your learning experience with a Lifetime Membership! Gain unlimited access to our vast resources and enjoy the added perk of receiving exciting new content every month.
Get Lifetime Access