Model

A8 model card

A8 on one page: model IDs, pricing, capabilities, endpoints, features, rate limits and how it compares.

Preview. Applies to a8-1. Last reviewed 2026-10-26.

This page as Markdown: https://u22a8.ai/docs/a8.md. The index of every docs page: https://u22a8.ai/llms.txt.

Model IDPreview

For judging text against criteria stated in plain words, at the accuracy each request sets.

Minimum accuracy
0–99, default 90
Context window
TBD (tokens)
Input price
$2 per 1M tokens
Latency
TBD (median)

Overview

A8 (Touchstone) is a decision model with a chat-model interface. It reads a conversation, judges its subject against the criterion in the system message and returns a verdict in the shape your schema gives. Each request sets a minimum accuracy: A8 answers where its evidence meets it and abstains everywhere else. A case it abstains on goes to a person. Tell your assistant their answer, and over MCP your organization’s model learns it in the background.

How it compares

AttributeTrained classifierLLM judgeDecision modelA8
The criterionTraining labelsIn the promptIn the questionIn the prompt
OutputA label or a scoreText or JSONA choice or a score, with a probabilityA verdict in JSON
AccuracyMeasured before useMeasured afterwardsA threshold you tune on labeled data, per versionSet per request, in the model’s tag; A8 finds the threshold
DeterminismDeterministicStochasticDeterministicDeterministic
What it starts fromA labelled dataset per criterionA promptA pretrained model, the same for every customerA pretrained model, and examples for your own criteria
Learning a new answerRelabel, retrain and redeployRewrite the promptReword the questionTell your assistant; it teaches A8
Why it answeredFeature weightsA written rationaleThe probabilityThe taught examples behind the verdict

Specifications

Model IDs

IdentifierPoints to
a8The latest version, at your organization’s latest ready horizon
a8:95The same model, at a minimum accuracy of 95
a8-1-1@20260715t140322zVersion 1.1 at one horizon: your data held at that moment
a8@20260715t140322zThe same model as the full form
a8-1-1@referenceVersion 1.1 as we publish it, before any teaching

Every response names the full identifier that answered, with the minimum accuracy it answered at as its tag. See Model identifiers and Horizons.

Pricing

MetricPriceUnit
Input$21M tokens
Input, reference model$21M tokens
Outputn/aA verdict writes no output tokens
Reasoning$01M reasoning tokens
Cached input$01M cached input tokens
Teach$01 teach
Retract$01 retract
Fitting a new version$01 fit
Precedents and recall$21M tokens

Prices of A8 1.1 Preview. See Pricing.

Capabilities

FeatureValue
Input → outputA conversation in text → a verdict: a label, a number or JSON
Context windowTBD (tokens)
ConversationsMulti-turn: A8 answers the latest user message
Minimum accuracy0 to 99 as the model’s tag, a8:95; 90 without one
ReasoningOn by default; effort none or disabled thinking turns it off
DeterminismThe same request on the same version returns the same verdict
LearningTaught examples, over MCP, fitted in the background
LatencyTBD (median and 95th percentile)
LanguagesTBD (languages measured)
Reference horizonTBD (the reference model’s horizon)

Availability

FeatureValue
StatusPreview
ReleasedTBD (release date)
RetirementOn or after TBD (date)
Data scopeYour organization’s model, trained on what your organization teaches

Endpoints

EndpointRouteSupport
Chat CompletionsPOST /v1/chat/completionsSupported
ResponsesPOST /v1/responsesSupported
Anthropic MessagesPOST /v1/messagesSupported
DecisionsPOST /v1/decisionsSupported
Stored responsesGET, DELETE /v1/responses/{id}Supported
Input itemsGET /v1/responses/{id}/input_itemsSupported
ModelsGET /v1/modelsSupported
MCPhttps://api.u22a8.ai/mcpSupported
Batch/v1/batchesNot supported
Embeddings/v1/embeddingsNot supported

Every route is on https://api.u22a8.ai. Each interface page lists the fields A8 reads: Chat Completions, Responses, Anthropic Messages, Decisions and MCP.

Features

FeatureSupport
Structured outputs (json_schema)Supported
A forced function call as the verdict’s shapeSupported
Multi-turn conversationsSupported
StreamingSupported
Reasoning on and offSupported, through effort or thinking
Content filteringSupported, in each API’s own slot
Teaching and retractingSupported, over MCP
Precedents and recallSupported, with models:explain
Pinned versionsSupported
The json_object response formatNot supported
More than one choice (n)Supported, each computed on its own
Stored responses (previous_response_id)Supported

Modalities

ModalitySupport
TextInput and output
ImageNot supported
AudioNot supported

Rate limits

TierRPMTPM
PreviewTBD (requests per minute)TBD (tokens per minute)

Limits apply per organization. The response headers that report them are on Rate limits.

Good to know

  • A prompt written for another judge works as it is. A8 finds the criterion, the subject and the answer format, such as GRADE: C, in a single message.
  • The model’s tag sets the minimum accuracy: a8:95 asks for 95. Words about certainty in a prompt are read as part of the criterion.
  • Reasoning is on by default and reads a request however it is laid out. With reasoning off, A8 reads the layout as written: the system message is the criterion and the latest user message the subject. Each interface page names its switch.
  • The system_fingerprint on each response moves when anything behind a verdict moves. See Reproducibility.
  • “What have I taught you about refunds?” returns your taught examples, word for word. See Precedents.
  • GET /v1/models lists each version your organization can call, with its status and whether it is active.

Evidence

qed-bench measured the scoring approach A8 is built on. On essay quality (ASAP 2.0), its rank correlation with trained human raters was 0.815; the strongest LLM judge on the panel, Claude Opus 4.7, reached 0.813. For your criteria, the minimum accuracy each request sets is the promise, and your own tests check it.

Next steps