iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
OpenAI’s Decisions API is a generally available endpoint for turning a defined piece of text, an image, a conversation, or JSON into a bounded answer your software can act on. Send shared evidence and one or more named questions to POST /v1/decisions; each question returns a typed result: a probability for a predicate, a selected value for a choice, or a probability-weighted score for an ordered rubric. The service is generally available as of October 22, 2026, and currently supports gpt-6-luna only.
Use it when the application must make a constrained decision such as routing a ticket, checking whether damage is visible, or assigning severity. Use the Responses API with Structured Outputs when you need an object that follows your own JSON schema, and function calling when the model should request a tool with arguments. Decisions is a focused decision layer, not a replacement for general response generation.
What is the Decisions API, and what changed in October 2026?
The Decisions API is a dedicated OpenAI endpoint for classification, routing, prioritization, and other bounded decisions. Instead of asking for free-form prose and parsing it yourself, you define the questions your application needs answered and receive typed answers that can drive deterministic application code.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →OpenAI announced a public beta on October 6, 2026, followed by general availability on October 22, 2026. The current release accepts requests at POST /v1/decisions and supports gpt-6-luna. Availability, model support, and regional features can change, so production deployments should verify the current API reference and changelog before pinning an implementation.
Popular on Amazon
- Read Before You Buy — No Video Output: These adapters support charging and USB 2.0 data transfer, but cannot transmit video signals. Except for standard USB webcams (which use USB data only), they are not compatible with HDMI/DisplayPort cables, video-capable USB-C hubs, or docking stations with video output.
- Convert USB-A Ports to USB-C: Designed to connect USB-C earphones, cables, flash drives, card readers, and other USB-C accessories to standard USB-A ports. Plug-and-play with no drivers or software required.
- Aluminum Alloy Housing: Built with a sturdy aluminum alloy shell that aids in heat dissipation and protects against daily wear and scratches. Designed to maintain a stable and secure connection.
- Compact & Travel-Friendly: The ultra-compact design allows the adapter to stay plugged into your device without blocking adjacent ports or adding bulk, reducing wear and tear on your original USB ports.
- 12-Month Warranty: Backed by a 12-month manufacturer warranty for peace of mind. Designed to meet strict quality control standards for reliable everyday performance.
A request contains three conceptual parts:
- Model: currently
gpt-6-luna. - Input: shared evidence supplied as text or an ordered conversation.
- Questions: one to 400 uniquely named predicate, choice, or score questions.
The response includes an answers array, along with the model and usage information. Give every question a unique name; the service echoes that name so your code can map an answer to the correct action without relying on array position.
How does a Decisions request and response work?
Think of the API as a typed decision layer around shared evidence. You provide the evidence once, then describe one or more bounded questions about it.
| Part | What you provide | Why it matters |
|---|---|---|
model |
gpt-6-luna |
Identifies the model that evaluates the evidence. |
input |
A text string or ordered messages | Holds the common context used by every question. |
questions |
One to 400 objects with a type, unique name, and instructions | Defines the exact decisions your application wants. |
answers (response) |
One answer associated with each question name | Lets application code choose the next action. |
usage ype (response) |
Token and cache accounting | Supports cost monitoring and capacity planning. |
Keep instructions operationally specific. “Classify this” leaves the allowed outcomes unclear; “Does the ticket describe a duplicate charge supported by the order?” defines a testable predicate. For a choice, list distinct values. For a score, state what each ordered level means.
When should I use a predicate, choice, or score?
Choose the question type based on the shape of the action that follows. A predicate answers whether a condition is present, a choice selects one category, and a score places evidence on an ordered rubric.
Predicate: a yes-or-no condition expressed as probability
Use a predicate for questions such as “Does this image show visible damage?” The result is a probability from 0 to 1 that the condition is true. Your application then chooses a threshold appropriate to the cost of a false positive or false negative.
A probability is not a guaranteed truth value. A warehouse workflow might automatically open an inspection task only when the probability exceeds a threshold selected from labeled examples; a low-confidence result can instead enter human review.
Choice: one value from an unordered set
Use a choice for routing or categorical classification, such as assigning a support ticket to billing, shipping, or technical support. Supply distinct allowed values and define what evidence belongs in each category. The answer contains the selected value plus probabilities and confidence information.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallBecause choices are unordered, “high,” “medium,” and “low” are merely labels in a choice question. If those labels represent increasing severity and you need a numeric interpretation, use a score instead.
Score: an ordered rubric with a weighted result
Use a score when levels have a meaningful order, such as severity from minor to critical. Define between two and ten ordered levels. The returned score is the probability-weighted average of the level indices, so it can fall between two named levels rather than matching one level exactly.
Also popular on Amazon
- 5-in-1 USB-C Hub: Experience comprehensive connectivity featuring a Power Delivery input, two USB-A 2.0 ports, a USB-A 3.0 port, and an HDMI port. (Note: The USB-C power delivery input port is only for connecting an external wall charger to power your laptop and cannot power peripheral devices.)
- 90W Pass-Through Charging: Achieve optimal charging with 90W pass-through power to your laptop, supported by a total input of 100W, with the hub reserving 10W for operational efficiency. (Note: Wall charger not included.)
- Quick Data Transfers: Accelerate your productivity with rapid data transfers using a high-speed 5Gbps USB 3.0 port and two 480Mbps USB 2.0 ports.
- 4K HDMI Display: Enhance your visual experience with a hub capable of delivering 4K resolution at 30Hz in both mirror and extend modes. Please note that this hub is compatible with MacBook (macOS 12 and newer), Windows 10 and 11, ChromeOS, and laptops equipped with DP Alt Mode and Power Delivery. Note: This device is not compatible with Linux.
- What You Get: Anker USB-C Hub (5-in-1, 4K HDMI), welcome guide, 18-month warranty, and our friendly customer service.
For example, if levels are indexed from 0 through 3, probabilities distributed between levels 1 and 2 can produce a score of 1.6. Preserve the individual level probabilities when the downstream policy needs to distinguish uncertainty from a firm classification.
What evidence can the API evaluate?
Shared input can be a text string or an ordered conversation. Supported material includes user text, developer or system text, assistant output or refusal, function calls, custom tool calls, and the matching tool outputs. This allows a decision to use the same conversation and tool history that informed an earlier model response.
User content can contain text, images, and an input_json object or array. Images may be supplied as public HTTP(S) URLs or inline base64 data URLs. File IDs are not supported, so an application that stores an image in an object store must either expose an accessible URL or encode the bytes as a data URL before sending the request.
When should an image belong to a question rather than shared input?
Put an image in a question’s instructions when that image is specific to that question. This is useful when every question compares a different candidate image with the same reference evidence. Keep each candidate in its own question and stay within the request’s image limits.
Put evidence in shared input when all questions should see exactly the same material. Shared input is also the portion that prompt caching can reuse, so keeping stable evidence there can reduce repeated input work.
How do I make the smallest working request?
Start with one independent predicate and a short, unambiguous instruction. The following JSON shows the minimal request shape described by the API:
{
"model": "gpt-6-luna",
"input": "Evaluate this support ticket and order evidence.",
"questions": [
{
"type": "predicate",
"name": "duplicate_charge",
"instructions": "Does the ticket describe a duplicate charge supported by the order?"
}
]
}- Send the JSON body to
POST /v1/decisionsusing your normal OpenAI API authentication and HTTPS client. - Read the returned
answersarray and locate the object whose name isduplicate_charge. - Apply an application-owned threshold or review rule to the returned probability.
- Log the input identifier, question name, selected action, model, and usage data needed for audit and cost analysis. Avoid logging sensitive evidence unless your data policy permits it.
In an OpenAI SDK, the equivalent operation is client.decisions.create. The reference response exposes answers, model, and usage; write your integration against named fields rather than assuming answer order.
How should I combine several decisions in one request?
Place independent questions that use the same evidence in one request. A single support ticket could simultaneously receive a duplicate-charge predicate, a department choice, and a severity score. This reuses shared context and keeps the resulting decisions tied to one evidence snapshot.
Requests support one through 400 questions. The upper range is intended for independent checks; it does not turn a sequence of dependent decisions into one atomic workflow.
Also popular on Amazon
- Sleek 7-in-1 USB-C Hub: Features an HDMI port, two USB-A 3.0 ports, and a USB-C data port, each providing 5Gbps transfer speeds. It also includes a USB-C PD input port for charging up to 100W and dual SD and TF card slots, all in a compact design.
- Flawless 4K@60Hz Video with HDMI: Delivers exceptional clarity and smoothness with its 4K@60Hz HDMI port, making it ideal for high-definition presentations and entertainment. (Note: Only the HDMI port supports video projection; the USB-C port is for data transfer only.)
- Double Up on Efficiency: The two USB-A 3.0 ports and a USB-C port support a fast 5Gbps data rate, significantly boosting your transfer speeds and improving productivity.
- Fast and Reliable 85W Charging: Offers high-capacity, speedy charging for laptops up to 85W, so you spend less time tethered to an outlet and more time being productive.
- What You Get: Anker USB-C Hub (7-in-1), welcome guide, 18-month warranty, and our friendly customer service.
What if one decision depends on an earlier answer?
Split dependent logic into multiple requests. Send the first request, inspect its result in application code, then construct a second request whose input or instructions incorporate that result. For example, first classify a ticket as billing or shipping; only after a billing result should the application ask a billing-specific severity question.
This two-stage design makes the dependency explicit and prevents later questions from silently using an answer that has not yet been established. It also lets you apply different thresholds, permissions, or human-review rules at each stage.
How is Decisions different from Responses Structured Outputs and function calling?
All three patterns can produce machine-readable results, but they expose different control surfaces and imply different workflow semantics.
| Decision axis | Decisions API | Responses with Structured Outputs | Function calling |
|---|---|---|---|
| Answer shape | Predicate probability, selected choice with probabilities/confidence, or ordered score | Generated object that conforms to an application-defined JSON schema | Tool-call name and arguments selected by the model |
| Control surface | Bounded question types, allowed choice values, and two-to-ten score levels | Your schema controls fields and data types | Your tool contract controls callable arguments |
| Best fit | Classification, routing, prioritization, and thresholded decisions | Extracting or generating a structured record | Asking the application to execute a tool or operation |
| Evidence model | Shared text, conversation/tool history, images, and native JSON; optional question-specific images | General response input with a schema-constrained output | General response input leading to a tool request |
| Workflow semantics | Independent questions together; dependent steps chained across calls | One generated structured response unless your code adds orchestration | Model proposes a call, then your application executes and returns its output |
| Cost and latency target | Focused decision calls, input-only published pricing, and cache support | General response-generation pricing and behavior | General response-generation pricing and tool orchestration |
| Operations | Decision usage reporting, regional processing, and eligibility-dependent data controls | Use the controls available for the selected Responses deployment | Use the controls available for the selected Responses deployment |
Choose Structured Outputs when the requirement is “return an object with these fields.” Choose function calling when the requirement is “request this operation with these arguments.” Choose Decisions when the requirement is “evaluate this bounded question and expose calibrated decision signals.”
How much does the Decisions API cost?
OpenAI’s published 2026 pricing is $0.10 per 1 million uncached input tokens and $0.01 per 1 million cached input tokens. Output tokens are not charged. These are published base figures from OpenAI for 2026, not a universal invoice estimate.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Regional-processing premiums and long-context multipliers can apply. Estimate a deployment’s actual bill using the pricing guidance for its region and context length, then validate the estimate against the usage values returned by real requests.
How does prompt caching reduce repeated input cost?
Decisions places a cache breakpoint at the end of shared input by default. The reusable prefix must contain at least 1,024 tokens. Keep stable evidence byte-for-byte identical across requests; changing whitespace, ordering, or wording can prevent a match.
Use prompt_cache_key to partition callers when different tenants or workflows should not share cache partitions. Use explicit cache breakpoints when you need exact control over which prefix is reusable. Cache read and write counts appear under usage.input_tokens_details and are already included in total input-token counts, so do not add them again when calculating usage.
How should I set thresholds and validate decisions?
Do not adopt a universal probability cutoff. Predicate probabilities and choice or score distributions are model estimates; the right threshold depends on the action’s error costs.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsAlso popular on Amazon
- Dual Converters, Infinite Potential：Includes 2× USB C male to USB A female adapters and 2× USB A male to USB C female adapters. Perfect for a wide range of uses—tablets with Bluetooth keyboards, expand USB ports on macbook, and more. Two different converters for all your daily needs
- Next-Level 10Gbps & 3A Charging: No more slow 480Mbps, this usb to usb c adapter has a transfer speed of up to 10Gbps, allowing you to do more transferring in less time. This usb adapter fits both USB A and USB C charger, supporting up to 3A fast charging
- Upgraded Exquisite Craftsmanship: With an aluminum alloy housing and metal connector, the usbc to usb adapter is extremely durable and sturdy. Rigorously tested to withstand more than 10,000 times of plugging and unplugging, ensuring long-lasting performance
- Broad Compatible: The usb c to usb adapter widely supports all USB C/ USB A devices like laptops, tablets, cellphones, car chargers, and phone chargers. Such as compatible with MacBook Pro/Air 2023/2022, Thunderbolt 4/3 Devices,Apple MagSafe Watch 9/8/7/SE/Ultra, iPad Pro 2022/2021, Samsung Galaxy S23/S20/S10, and iPhone 17/16/15 Pro. Plug and play
- Please Note: To reach 10Gbps speed, keep the cable under 3.3 ft. For USB A Male to USB C adapters, try flipping the USB C connector. USB C Male to USB A adapters support bidirectional 10Gbps transfer within 3.3 ft
- Collect labeled examples representative of the traffic, including difficult and ambiguous cases.
- Run the exact question wording and evidence format you intend to deploy.
- Measure false positives and false negatives separately for each action.
- Choose a threshold that reflects the business cost of each error, not just the largest overall accuracy.
- Define a review band for uncertain results and monitor it after launch.
For a score, decide whether automation should use the weighted numeric score, the highest-probability level, or the full distribution. Those choices produce different behavior near level boundaries. Preserve the model probabilities in internal telemetry so you can recalibrate without reconstructing the original request.
What throughput and usage controls are available?
Use the one-to-400-question limit to consolidate independent checks over shared evidence, but split large or dependent workflows when they become difficult to audit. Keep question names stable across releases so dashboards and action logs remain comparable.
Organization administrators can query GET /v1/organization/usage/decisions and filter or group usage by project, API key, or model. The api.usage.read permission is required. Monitor request counts, input tokens, cache reads, and cache writes rather than relying only on a monthly invoice.
Can Decisions run in voice and regulated-data workflows?
The general-availability release adds native GPT-Live integration. A voice application can call Decisions directly or trigger a decision as transcript deltas arrive, which is useful for routing or intervention during a live interaction.
Free tools Windows power users keep installed
One-click scans. No signup required.
Regional processing is supported in the United States and Europe, including the European Economic Area and Switzerland. Zero Data Retention and HIPAA support are available to eligible customers subject to OpenAI’s data-control requirements and agreements; eligibility is not automatic, so confirm the applicable contract and project configuration before sending regulated data.
OpenAI quoted Andrew Hsu, CTO of Speak, saying: “In our testing, the Decisions API integration in GPT‑Live was able to reduce the time to next action by over 90%, from 1.4s to 0.13s, creating a much more interactive learning experience.” This is a named vendor statement reported by OpenAI in 2026, not an independent benchmark or a guarantee for other applications.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What can go wrong, and how do I fix it?
Answers are mapped to the wrong action
Usually the application is using array position instead of the echoed question name. Give every question a unique, stable name and map answers by that name.
A choice result is inconsistent between similar cases
Check that allowed values are distinct and that each instruction defines the boundary between categories. Add representative labeled examples to your evaluation set, then revise the category definitions rather than silently treating a low-confidence result as certain.
Recommended Free Tools
Scores appear between named levels
That is expected: score is a probability-weighted average of ordered level indices. If a downstream system accepts only discrete labels, define an explicit rounding or banding policy in application code and test it at every boundary.
Also popular on Amazon
- 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
- Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
- Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
- HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
- What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.
Cache usage stays at zero
Verify that the shared input is at least 1,024 tokens, identical across requests, and placed in the shared-input portion rather than rewritten inside each question. Check usage.input_tokens_details to distinguish a cache miss from a request that did not contain a reusable prefix.
A dependent question produces an unjustified answer
Do not place dependent questions beside the first question and assume ordering creates a dependency. Make the first call, validate its result, and send a second request with the relevant result explicitly included.
An image cannot be retrieved
Use a publicly reachable HTTP(S) URL or an inline base64 data URL. File IDs are not supported by this endpoint, and a URL that requires a private login will not provide usable evidence to the service.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
What should I check before production?
- Confirm the endpoint, model support, and regional availability for the deployment date.
- Give every question a unique name and map answers by name.
- Use predicate, choice, and score only when their semantics match the action.
- Keep independent questions together and chain dependent decisions across requests.
- Define thresholds from labeled examples and document false-positive and false-negative costs.
- Keep stable shared evidence byte-for-byte identical when using prompt caching.
- Track cache details, token usage, request counts, and model identifiers.
- Test public image URLs, inline data URLs, JSON inputs, conversation history, and tool outputs in the formats your application will actually send.
- Confirm data-retention, regional-processing, and HIPAA requirements with the applicable OpenAI agreement before handling regulated information.
- Provide a human-review path for low-confidence or high-impact decisions.
Is the Decisions API the right choice for my application?
Use Decisions when your application needs a small, explicit set of decisions over shared evidence and can define what each result means operationally. Its typed probabilities, categorical values, ordered scores, multimodal input, caching, and usage reporting reduce the parsing and orchestration work common with free-form model output.
Keep using Responses Structured Outputs for schema-shaped extraction or generation, and function calling for model-initiated tool execution. Treat Decisions probabilities as signals to calibrate and monitor, not as universal truth scores; the production quality of the system depends on your labels, thresholds, review policy, and action design.
Frequently Asked Questions
Can I use a private image URL with the Decisions API?
No. Image evidence must be supplied through a publicly reachable HTTP(S) URL or an inline base64 data URL; a URL that requires an application login will not work as evidence. If exposing an image publicly is unacceptable, encode it inline and apply the data-handling controls required by your organization.
How should I evaluate a new question before enabling automation?
Create a holdout set that was not used to write the instructions, run the exact production request shape, and inspect errors by category and confidence range. Select an automation threshold only after comparing the operational cost of false positives, false negatives, and human review.
Will a model update preserve my existing probability thresholds?
Do not assume it will. Re-run a representative labeled set after a model or prompt change, compare the probability distributions and action-level error rates, and re-approve thresholds before allowing the updated workflow to act automatically.
Can one request contain unrelated customer cases?
It is safer to keep each request’s shared input tied to one evidence unit, such as one ticket or conversation. Mixing unrelated cases makes provenance, caching boundaries, and answer-to-action auditing harder even when the question limit permits many questions.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
