- Published on
Llm-openai-decisions 0.1a0
- Authors
- Name
- DeepHorus
OpenAI released its new Jev-style Decisions API on 6th October 2026, delivering on the capability first previewed during the company's DevDay keynote the previous week. The launch marks OpenAI's formal entry into a category of structured decision-making endpoints pioneered by Jev, a competing provider of similar tooling. The API lets developers submit questions and receive machine-readable verdicts rather than free-form prose. It targets use cases where applications need reliable, structured judgments instead of open-ended generation.
The Decisions API is designed to be conceptually familiar to anyone who has worked with Jev's offering. Like Jev, it supports three question types: yes/no predicates, multiple-choice selections, and numeric scores. The prompt for building a companion tool came from an existing llm-typesafe plugin used for communicating with Jev, which served as the structural blueprint. Rather than hand-writing the integration, the developer had GPT-6 Astra read the freshly published OpenAI API documentation and generate the new plugin from scratch.
The result is llm-openai-decisions, an open-source plugin installable with a single command: llm install llm-openai-decisions. It exposes a decision model identified as gpt-6-luna, which differs from its Jev counterpart by accepting image input in addition to text. An illustrative query attaching a photograph of two pelicans and asking whether the image contains any mammals returns a structured response: {"type": "predicate", "name": "evaluation", "probability": 0.0}. Full instructions for running the other two question types are documented in the plugin's README.
Pricing distinguishes the two rivals clearly. OpenAI charges 10 cents per million input tokens for gpt-6-luna, while Jev charges 4.2 cents per million for its equivalent. Neither provider bills for output tokens, a notable departure from conventional LLM pricing models that charge for generated text. That structure rewards applications producing large, verbose responses and keeps costs predictable for integrators.
The practical significance extends beyond pricing. Structured decision endpoints reduce the parsing errors and hallucinated formatting that plague prompt-engineered JSON extraction, giving applications deterministic schemas to consume. For teams already standardized on Jev, the near-identical API shape means migration requires minimal code changes. The ability to attach images also broadens the range of questions that can be answered programmatically, from content moderation to document verification.
The broader ecosystem context matters too: OpenAI's DevDay 2026 took place on 29th September 2026, and the Decisions API was among its headline announcements. Developers now face a genuine two-vendor choice on price, model capabilities, and ecosystem maturity. Whether OpenAI's larger distribution footprint outweighs Jev's pricing advantage will likely determine adoption patterns in the coming months.
Looking ahead, additional plugins and wrappers are expected as the documentation matures. Analysts have also noted that structured decision APIs may eventually require hard budget caps, a theme raised in commentary published on 3rd October 2026.
