Skip to main content

Aion Labs Provider Guide

Aion Labs is a Tier-2 catalog provider: OpenAI-wire-compatible with no behavioural quirks, so its entire integration is one JSON file (src/lib/providers/catalog/aionlabs.json) rather than hand-written code. That file is the source of truth for everything on this page.

Verification status: this entry is docs- and roster-verified, not yet live-verified. Every field below comes from Aion Labs' own public documentation and an unauthenticated GET /v1/models call — no account was created and no API key was used to build it. evidence.liveMatrix is null until someone runs the live capability matrix with a real key (see Verification status below).

Key Facts​

  • Provider id: aionlabs
  • Protocol: OpenAI-compatible (/chat/completions) — the docs say the API is "Drop-in compatible with OpenAI's Chat API" (https://www.aionlabs.ai/docs/)
  • Base URL: https://api.aionlabs.ai/v1
  • Default model: aion-labs/aion-3.5
  • Models in catalog: 6 (the whole roster — the live GET /v1/models call returned exactly 6, all text->text)
  • Model focus: the docs models page (https://www.aionlabs.ai/docs/models/) describes Aion 2.0 and Aion-RP as "optimized for immersive roleplaying and storytelling", and the other four (Aion 3.0, 3.0 Mini, 3.5, 3.5 Mini) as giving stronger narrative structure, more compelling tension and conflict, and a nuanced treatment of mature and darker themes. The homepage (https://www.aionlabs.ai/) describes the API as for creative writing, roleplay, research and production workloads. None is described as vision-capable.
  • Streaming: supported — the API reference documents stream: true with text/event-stream deltas ending in data: [DONE]
  • Tool calling: model-dependent — the API reference documents tools "in OpenAI format", tool_calls and role: tool messages, but has no per-model support table, so nothing says every model handles them
  • Structured output: not declared (false) — the request-body table in the API reference lists no response_format parameter. generate({ schema }) still returns parsed structuredData through NeuroLink's client-side JSON coercion.
  • Structured output + tools together: not declared (false) — no combined probe was possible without credentials
  • Tools + streaming: false — not live-verified. The API reference (https://www.aionlabs.ai/docs/api-reference/) describes POST /v1/chat/completions as "OpenAI-compatible chat completions. Supports streaming and tool calls."
  • Thinking: not declared (false) — five of the six models reason and the API accepts a reasoning_effort parameter, but the generic catalog provider does not send it, so NeuroLink's thinkingLevel is not mapped onto it (see Reasoning models)
  • Embeddings: not declared — the docs list only chat completions, /v1/responses and model discovery
  • Billing: free-tier — the public pricing page offers a Free Tier with "A daily credit allowance … No card required."
  • Key format: none declared by Aion Labs

Quick Start​

1. Get an API key​

  1. Visit: https://www.aionlabs.ai/accounts/signup/ and sign up (GitHub, Google or email)
  2. Create an API key from the dashboard's API keys page (https://www.aionlabs.ai/app/api-keys/, sign-in required) and copy it when it is shown
  3. Free tier: the pricing page (https://www.aionlabs.ai/pricing/) offers a daily credit allowance with no card required; free-tier limits are 15 requests/min, 20,000 tokens/min and 20,000 tokens/day (https://www.aionlabs.ai/docs/rate-limits/)
  4. Set AIONLABS_API_KEY in your .env file

Whether a brand-new free account can create a key immediately, or waits on manual approval (the site header also carries a "Request Access" button), was not tested — no signup was performed to build this entry.

2. Configure​

export AIONLABS_API_KEY=your-api-key
export AIONLABS_MODEL=aion-labs/aion-3.5   # optional — overrides the default model
export AIONLABS_BASE_URL=https://api.aionlabs.ai/v1   # optional — proxy or self-hosted gateway

3. Use it​

import { NeuroLink } from "@juspay/neurolink";

const neurolink = new NeuroLink();

const result = await neurolink.generate({
  input: { text: "Write the opening paragraph of a lighthouse mystery." },
  provider: "aionlabs",
  model: "aion-labs/aion-3.5",
  maxTokens: 4096,
});

console.log(result.content);
# CLI
npx @juspay/neurolink generate "Hello" --provider aionlabs --max-tokens 4096

Per-request credentials work as they do for every provider:

await neurolink.generate({
  input: { text: "Hello" },
  provider: "aionlabs",
  credentials: { aionlabs: { apiKey: process.env.AIONLABS_API_KEY } },
});

Models​

ModelContextVision$/M in · outNotes
aion-labs/aion-3.5 ⭐262,144no$3.00 / $6.00 (cached $0.75)Recommended — the model every request example on Aion Labs' quickstart uses; released 2026-09-23, the latest release date on the roster (shared with 3.5 Mini); GLM-family multi-model system.
aion-labs/aion-3.5-mini262,144no$0.70 / $1.40 (cached $0.18)Released 2026-09-23; GLM-family multi-model system; roster-listed only; fallback.
aion-labs/aion-3.0131,072no$3.00 / $6.00 (cached $0.75)Released 2026-05-05; GLM-family multi-model system; roster-listed only.
aion-labs/aion-3.0-mini131,072no$0.70 / $1.40 (cached $0.18)Released 2026-05-14; DeepSeek-family multi-model system; roster-listed only; fallback.
aion-labs/aion-2.0131,072no$0.80 / $1.60 (cached $0.20)Variant of DeepSeek V3.2 for roleplay and storytelling (2025-12-21); roster-listed only; fallback.
aion-labs/aion-rp-llama-3.1-8b32,768no$0.80 / $1.60Aion-RP 1.0 (8B), a Llama 3.1 8B variant (2024-11-30); the only non-reasoning model on the roster; roster-listed only.

Every model has a maximum completion of 32,768 tokens. Context comes from the context_length field of an unauthenticated GET https://api.aionlabs.ai/v1/models call made 2026-09-29 (6 models; the docs models page at https://www.aionlabs.ai/docs/models/ rounds the same values to 128K / 256K / 32K). Prices are the per-million-token figures on that docs models page, retrieved the same day; the roster's per-token pricing fields agree with them. models.defaultContextWindow (262,144) and models.defaultMaxOutputTokens (32,768) are the default model's own figures. The docs pricing page (https://www.aionlabs.ai/docs/pricing/) served the same table as the models page when it was fetched.

Fallback order when the default is unavailable: aion-labs/aion-3.5-mini → aion-labs/aion-3.0-mini → aion-labs/aion-2.0.

vision is false on every model because the roster reports text->text modality for all six; no image-input request was sent.

Reasoning models​

The five aion-3.x and aion-2.0 models are tagged "Reasoning" on the docs models page and reasoning: true on the roster; aion-rp-llama-3.1-8b is not. The API reference (https://www.aionlabs.ai/docs/api-reference/) documents two behaviours that matter when you call them through NeuroLink:

  • Reasoning shares the max_tokens budget. Reasoning is generated first, so a cap exhausted during reasoning returns finish_reason=length with empty content. Give these models a generous maxTokens; it is a ceiling, not a reservation, and unused budget is not billed.
  • reasoning_effort differs by generation. aion-3.5 and aion-3.5-mini accept exactly low, high (default) or max and cannot have reasoning disabled — any other value returns a 400. aion-2.0, aion-3.0 and aion-3.0-mini also accept none and medium (default medium).

NeuroLink's catalog provider does not send reasoning_effort, so the vendor defaults apply.

Verification status​

Tier-2 onboarding requires evidence before a provider is accepted, and pnpm run verify:provider-onboarding gates it in CI. This is what the catalog currently records for Aion Labs — docs- and roster-verified, not live-verified:

ProbeResult
Rosterunauthenticated GET /v1/models, HTTP 200, 6 models, 2026-09-29 — no key used
Billingpublic pricing page states a Free Tier ("A daily credit allowance … No card required."); the rate-limits page lists Free as the default tier on signup, 2026-09-29. No signup was performed, so key issuance for a fresh account is unconfirmed
Auth-failure shapenot documented — the API reference's error table lists only 400 invalid_request_error, 429 rate_limit_error and 502 server_error, and no credentialed request was made. The only errorRules entry is the documented 429; 401s use the default classifier
Tools / structured outputtools are documented on the API reference but not per model and were not exercised live; no response_format parameter is documented, so structuredOutput is false; no combined tools+schema request was sent — structuredOutputWithTools stays false
Live capability sweepnot run. evidence.liveMatrix is null. Before treating this provider as production-ready, run npx tsx test/continuous-test-suite-provider-matrix.ts --provider=aionlabs with a real key and record the result.

Do not treat this entry as equivalent to a live-verified Tier-2 provider (e.g. FriendliAI, Novita AI) until that live matrix has been run and evidence.liveMatrix is filled in.

Troubleshooting​

SymptomCauseFix
Empty content with finish_reason=lengthThe max_tokens cap was used up by reasoning before any visible answerRaise maxTokens; on the 3.5 models reasoning cannot be turned off
HTTP 429 rate_limit_errorUsage limit exceeded — free tier is 15 requests/min, 20,000 tokens/min and 20,000 tokens/dayWait, or top up credits to reach a higher tier (https://www.aionlabs.ai/docs/rate-limits/)
HTTP 524 on a long requestNon-streaming responses must complete within 125 seconds; generation still finishes and is billedUse streaming for large maxTokens, reasoning models and very long prompts
HTTP 400 invalid_request_errorUnknown model or malformed request; an unsupported reasoning_effort value also returns 400Pick a current id from the unauthenticated GET /v1/models roster
HTTP 502 server_errorTemporary server-side errorRetry the request
Structured output ignored with tools attachedstructuredOutputWithTools is false on this entry — untested combinationNeuroLink omits response_format automatically whenever tools are present, before sending
Tool calls not returnedThe API reference says the endpoint "Supports streaming and tool calls" but gives no per-model tool support, and none was testedUnverified per model — check the request against the OpenAI tools format the API reference describes

See also​