Model Registry

The library includes registry schema v2 metadata for the current text/realtime catalog, plus separate TTS, STT, image, video, and embedding registries.

Using the Model Registry

typescript

import {
  getModelInfo,
  calculateCost,
  getModelsByVendor,
  getActiveModels,
  getDeprecatedModels,
  LLM_MODELS,
  MODEL_REGISTRY_SCHEMA_VERSION,
  Vendor,
} from 'oneringai';

// Get model information
const model = getModelInfo('gpt-6-astra');
if (!model) throw new Error('Model is not registered');
console.log(model.provider);                   // 'openai'
console.log(MODEL_REGISTRY_SCHEMA_VERSION);   // 2
console.log(model.features.input.tokens);     // 1050000 total context-window guardrail
console.log(model.features.output.tokens);    // 128000
console.log(model.features.reasoning);        // true
console.log(model.features.vision);           // true
console.log(model.lifecycle);                  // 'active'
console.log(model.endpoints);                  // ['responses', 'chat_completions', 'batch']

// Calculate API costs
const cost = calculateCost('gpt-5.6-luna', 50000, 2000);
console.log(`Cost: $${cost}`); // $0.0124

// With mixed cached and uncached input, using batch pricing
const cachedCost = calculateCost('gpt-5.6-luna', 50000, 2000, {
  cachedInputTokens: 30000,
  processingMode: 'batch',
});
console.log(`Optimized: $${cachedCost}`); // $0.0035

// Get all models for a vendor
const openaiModels = getModelsByVendor(Vendor.OpenAI);
console.log(openaiModels.map(m => m.name));
// ['gpt-6-astra', 'gpt-5.6-sol', 'gpt-5.6-terra', 'gpt-5.6-luna', ...]

// Get all active models
const activeModels = getActiveModels();
console.log(activeModels.length); // lifecycle-aware active records only

// Use model constants
const flagship = LLM_MODELS[Vendor.OpenAI].GPT_6_ASTRA;
const efficient = LLM_MODELS[Vendor.OpenAI].GPT_5_6_LUNA;

Registry schema v2 and API compatibility

Schema v2 is additive for existing consumers. name, provider, isActive, features, and the original cost fields retain their meaning. New metadata is:

lifecycle: preview, active, legacy, deprecated, or retired;
availability, preferred, aliases, and snapshots;
endpoints, deprecationDate, retirementDate, and replacementModel;
official sources on media records and newly audited text entries; and
long-context, processing-mode, duration, message, resolution, and modality pricing where the vendor publishes it.

Unknown token limits are represented as null, not 0. This currently matters for duration-priced voice-agent models. Existing code that assumes a number should use resolveMaxContextTokens() or check for null.

Lookup helpers resolve aliases. Direct indexing remains canonical-only:

typescript

getModelInfo('gpt-5.6')?.name;          // 'gpt-5.6-sol'
getModelInfo('grok-voice-latest')?.name; // 'grok-voice-think-fast-2.0'
MODEL_REGISTRY['gpt-5.6'];              // undefined by design

Use getDeprecatedModels() and the equivalent media helpers instead of treating every callable-but-deprecated model as inactive. isActive means the API can still be called; lifecycle carries the vendor's migration state.

Current provider API selection

OpenAI text models use Responses by default, including current reasoning, service-tier, explicit/implicit prompt-cache, hosted-tool, and long-context cost metadata. Connector-first wrappers also expose Decisions, GPT-Live, and custom voices.
Anthropic supports Claude 5.5 adaptive/between-tools thinking, effort through output_config, Opus fast mode, tool search/deferred tools, server-side compaction, and current server tools. Generic request metadata is narrowed to Anthropic's supported { user_id } shape; unrelated host keys remain local rather than producing invalid Messages requests.
Gemini 3.5+ uses the GA Interactions API (steps, step.delta) by default; Gemini 3.8 also supports continuation tokens, file search, and computer use. Set vendorOptions.api = 'generateContent' to retain the legacy path, or vendorOptions.api = 'interactions' to force Interactions on another model. Named function selection is strict on both paths (allowed_tools for Interactions and allowedFunctionNames for generateContent). Live sessions and custom voices have connector-first wrappers.
xAI text uses a dedicated OpenAI-compatible adapter for Grok reasoning, registry-gated X/web search, hosted code execution, ordered replay, and cost telemetry; dedicated image, video, TTS, STT, and realtime voice endpoints use native contracts.

Resolve Model Capabilities

Providers use the registry to resolve model capabilities automatically, but you can also use the helper directly:

typescript

import { resolveModelCapabilities, resolveMaxContextTokens } from 'oneringai';

// Get full capabilities for a registered model
const caps = resolveModelCapabilities('gpt-5.6-terra', {
  supportsTools: true, supportsVision: false, supportsJSON: true,
  supportsJSONSchema: false, maxTokens: 128000, maxOutputTokens: 4096,
});
console.log(caps.maxTokens);       // 1050000 (from registry)
console.log(caps.maxOutputTokens); // 128000 (from registry)

// For unregistered models, vendor defaults are returned
const unknown = resolveModelCapabilities('my-custom-model', { ...vendorDefaults });
console.log(unknown.maxTokens);    // falls back to vendorDefaults

// Quick context limit lookup (useful for error messages)
const limit = resolveMaxContextTokens('gpt-5.6-terra', 128000);
console.log(limit); // 1050000

Model Information

typescript

interface ILLMDescription {
  name: string;
  provider: string;
  description?: string;
  isActive: boolean; // callable, not necessarily recommended
  lifecycle?: 'preview' | 'active' | 'legacy' | 'deprecated' | 'retired';
  availability?: 'public' | 'limited' | 'invite_only' | 'enterprise' | 'region_limited';
  preferred?: boolean;
  aliases?: readonly string[];
  snapshots?: readonly string[];
  endpoints?: readonly ModelEndpoint[];
  releaseDate?: string;
  deprecationDate?: string;
  retirementDate?: string;
  replacementModel?: string;
  knowledgeCutoff?: string;
  sources?: {
    documentation: string;
    pricing?: string;
    apiReference?: string;
    lastVerified: string;
  };

  features: {
    input: {
      tokens: number | null;
      cpm: number;
      cpmCached?: number;
    };
    output: {
      tokens: number | null;
      cpm: number;
    };

    // Feature flags
    reasoning: boolean;
    streaming: boolean;
    structuredOutput: boolean;
    functionCalling: boolean;
    vision: boolean;
    audio: boolean;
    video: boolean;
    extendedThinking: boolean;
    batchAPI: boolean;
    promptCaching: boolean;

    pricing?: {
      text?: TokenPricing;
      audio?: TokenPricing;
      image?: TokenPricing;
      audioDurationPerMinute?: number;
      textInputPerMessage?: number;
      processingMultipliers?: Partial<Record<ProcessingMode, number>>;
    };
  };
}

Available Models

OpenAI:

GPT-6: Astra, GPT-6.1 Sol, GPT-6 Sol, and GPT-6 Luna
Specialized: GPT-5.6 Cyber and GPT-Rosalind Research (access-controlled)
GPT-5.6: Sol, Terra, Luna
GPT-5.5 Pro
GPT-5.3: codex, chat-latest
GPT-5.2: standard, pro, codex, chat-latest
GPT-5.1: standard, codex, codex-max, codex-mini, chat-latest
GPT-5: standard, mini, nano, codex, chat-latest
GPT-4.1: standard, mini, nano
GPT-4o: standard, mini
Audio: gpt-audio-1.5, gpt-audio, gpt-audio-mini
Realtime: gpt-realtime-2.1 (recommended), gpt-realtime-2.1-mini, gpt-realtime-2, gpt-realtime-translate, gpt-realtime-1.5; gpt-realtime and gpt-realtime-mini are deprecated
Realtime transcription: gpt-live-transcribe (recommended) and gpt-realtime-whisper ($0.017/audio minute)
o-series: o3, o4-mini, o3-mini, o1
Deep Research: o3-deep-research, o4-mini-deep-research
Open-weight: gpt-oss-120b, gpt-oss-20b

Anthropic:

Claude 5.5: Opus 5.5 and Sonnet 5.5
Claude 5: Opus 5, Mythos 5, Fable 5, and Sonnet 5
Opus 4.8
Claude 4.7: Opus 4.7 (legacy flagship)
Claude 4.6: Opus 4.6, Sonnet 4.6
Claude 4.5: Opus, Sonnet, Haiku
Claude 4.x legacy: Opus 4.1, Opus 4, Sonnet 4
Claude 3.7: Sonnet

Google:

Gemini 3.8 Flash, Live, and Live Extended Thinking
Gemini 3.7 and 3.6 Flash; Gemini 3.5 and 3.5 Flash-Lite
Gemini 3.1: Pro, Flash-Lite, Flash Image, Flash Live
Gemini 3 (preview): Flash, Pro Image
Gemini 2.5: Pro, Flash, Flash-Lite, Flash Image

Grok / xAI:

Grok 4.7, Grok 4.6, Grok 4.5, Grok 4.3, and Grok Build 0.1
Grok 4.20 (2M context, flagship): reasoning, non-reasoning, multi-agent
Grok 4.1 fast (2M context): reasoning, non-reasoning
Grok Voice Think Fast 2.0 and maintained 1.0 migration entries

DeepSeek (5 models):

V4 Flash, V4 Pro, and V4 Flash Vision Experimental
Retired deepseek-chat and deepseek-reasoner compatibility records

Dedicated media registries add GPT Image 2.5 Sunburst/Flare, Gemini Nano Banana 2.1 and 3.1 Flash-Lite Image, Grok Imagine Image 2.0, Sora 2, Veo/Omni, Grok Imagine Video 1.5, current OpenAI speech, Gemini 3.8 TTS, Grok Voice Transcribe 2.0, Google/xAI voice models, and Gemini Embedding 2. Use getImageModelInfo, getVideoModelInfo, getTTSModelInfo, getSTTModelInfo, and getEmbeddingModelInfo for their modality-specific capability schemas.

The complete official-source audit and migration table is in docs/MODEL_REGISTRY_AUDIT.md.