Decision models offer a cheaper path for bounded questions
OpenAI’s generally available Decisions API, Baseten’s downloadable decision models and Perplexity’s local Mac release give developers more ways to classify or score inputs without generating a long answer.
OpenAI made its Decisions API generally available on October 22 with GPT-6 Luna.
Developers supply text, images or structured data and ask for bounded outputs such as a choice, score or probability. The release supports up to 400 questions per request and caching of shared context. The published standard price starts at $0.10 per million uncached input tokens, with no output-token charge; caching, long contexts and regional processing have their own conditions. [1][2]
Baseten’s October 21 Mammoth-1 and Elephant-1 releases provide a downloadable alternative. Built from Qwen3.5 and sized at 27 billion and 9 billion parameters, they target decisions such as boolean checks, selections and scoring. Both model repositories identify the MIT licence. Their task-specific design is the relevant distinction from a general chatbot; Baseten’s claimed benchmark lead does not establish superiority on every business rule. [3][4][5]
Take it further
Check whether Decider fits your Mac Public setup guide; Apple Silicon and stated memory, OS and disk requirements apply.
Sources & details 8 sources
Discuss in your voice app
Take the reporting, visual explanations and original sources into a conversation. Copy the brief, paste it into your preferred assistant, then start its voice mode.
Paste the brief into your voice app to discuss this reporting.Reporting sources
- OpenAI — Decisions API general availability
October 22 release; output types, inputs and availability.
- OpenAI API — Decisions guide
Current pricing conditions, batching and probability-calibration guidance.
- Baseten — Mammoth-1 and Elephant-1
October 21 release; vendor benchmark claims are not generalized.
- Baseten — Mammoth-1 model card
Model provenance and MIT licence.
- Baseten — Elephant-1 model card
Model provenance and MIT licence.
- Perplexity — local Mac release
October 23 official community announcement.
- Perplexity — Decider for Mac setup
Hardware, operating-system and download requirements; local execution.
- Calibration methodology
Guo et al. (ICML 2017), background on confidence calibration; not an evaluation of the newly released decision models.
Images & figures
A finite answer set feeds an application rule that routes uncertain cases for review.
Euangelion / Codex · Source · Usage terms