Skip to main content

Meta Model API Provider Guide

Meta Model API is a Tier-2 catalog provider: OpenAI-wire-compatible, so its entire integration is one JSON file (src/lib/providers/catalog/meta-model-api.json) rather than hand-written code. That file is the source of truth for everything on this page.

Verification status: this entry is docs-verified only, not yet live-verified. The vendor's GET https://api.meta.ai/v1/models needs a key (an unauthenticated GET on 2026-09-29 answered HTTP 401), so the model ids come from Meta's public docs and the roster has not been checked. No account was created and no API key was used to build this entry. evidence.liveMatrix is null until someone runs the live capability matrix with a real key (see Verification status below).

Key Facts​

  • Provider id: meta-model-api
  • Protocol: OpenAI-compatible (/chat/completions). The Chat Completions page says "The endpoint is OpenAI-compatible" and "The response matches the OpenAI chat completions shape."
  • Base URL: https://api.meta.ai/v1 — the Base URL in the At a glance block of the Overview, and the base_url value in the Chat Completions page's OpenAI SDK examples
  • Default model: muse-spark-1.3 — the Models page says "muse-spark-1.3 is the default model in the code examples throughout these docs."
  • Models in catalog: 5 — the five rows of the "Available Muse Spark models" table on the Models page (three Standard tier, two Contributor tier)
  • Streaming: supported — the Chat Completions page says "Set stream=True to enable streaming:"
  • Tool calling: true — the Tool calling page says "tool calling is fully supported there too" for Chat Completions
  • Tools while streaming: true — the Tool calling page says "Tool-call arguments stream: With stream: true, the model streams arguments incrementally."
  • Structured output: true — the Structured output page says "Set response_format to type: "json_schema" and provide your schema in the json_schema field."
  • Structured output + tools together: false — no combined probe was run without credentials
  • Embeddings: false
  • Thinking: false — the entry sets no thinking-level request parameter. The Chat Completions page says "Muse Spark always reasons, so reasoning_effort: "none" returns HTTP 400." The Reasoning page documents the request parameter: "On Chat Completions, use the top-level reasoning_effort parameter."
  • Billing: schema value no-free-tier, a placeholder; see Billing below
  • Key format: LLM|{id}|{secret} — the Error handling page lists Malformed API key (not in LLM|{id}|{secret} format) as a cause of HTTP 401; the entry's apiKeyFormat is ^LLM\|.+\|.+$

Quick Start​

1. Get an API key​

  1. Visit: https://dev.meta.ai/ — the page carries a "Start building" button whose link target is /api/auth/login
  2. Create an API key: the Authentication page (https://dev.meta.ai/docs/authentication) lists the steps "Log in and open the API keys tab.", "Click Create API key, give it a descriptive name, and click Create." and "Copy the key right away. You only see it once."
  3. Authentication is a Bearer token: the Authentication page says "Pass the key as a Bearer token in the Authorization header:", and the Error handling page (https://dev.meta.ai/docs/error-handling) lists Malformed API key (not in LLM|{id}|{secret} format) as a cause of HTTP 401
  4. Billing as the vendor states it: the Pricing page (https://dev.meta.ai/docs/pricing-rate-limits) says "You pay only for what you use." and the Authentication and billing page (https://dev.meta.ai/docs/muse-code/auth) says "Meta Model API billing is usage-based: you're billed for the tokens your requests consume."
  5. Set META_MODEL_API_API_KEY in your .env file (the vendor's examples read MODEL_API_KEY, which NeuroLink does not)

2. Configure​

export META_MODEL_API_API_KEY=your-api-key
export META_MODEL_API_MODEL=muse-spark-1.3   # optional — overrides the default model
export META_MODEL_API_BASE_URL=https://api.meta.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: "Explain context windows in one paragraph." },
  provider: "meta-model-api",
  model: "muse-spark-1.3",
});

console.log(result.content);
# CLI
npx @juspay/neurolink generate "Hello" --provider meta-model-api

Per-request credentials use the same call shape as the other catalog providers:

await neurolink.generate({
  input: { text: "Hello" },
  provider: "meta-model-api",
  credentials: { metaModelApi: { apiKey: process.env.META_MODEL_API_API_KEY } },
});

Billing​

The Pricing page (https://dev.meta.ai/docs/pricing-rate-limits) says "You pay only for what you use." and lists these prices per 1M tokens:

TierCached inputInputOutput
Standard$0.15$1.25$4.25
Contributor$0.002$0.10$0.20

The Authentication and billing page (https://dev.meta.ai/docs/muse-code/auth) says "Meta Model API billing is usage-based: you're billed for the tokens your requests consume." and lists two steps for a team admin: "Add a payment method." and "Create an API key." The Pricing page's Muse Voice Transcribe paragraph says "ZDR is priced at parity with Standard, and platform free-tier credits apply."

The entry records the schema value no-free-tier as a placeholder: the schema has no unknown value, and the value is not a vendor statement.

Rate limits on the Pricing page: Standard tier 3,000 requests per minute and 4,000,000 tokens per minute; Contributor tier 100 requests per minute and 3,000,000 tokens per minute. The page says "Limits apply per team, not per API key."

Models​

ModelTierContextVision$/M in · out (cached input)Notes
muse-spark-1.3 ⭐Standard1,048,576yes$1.25 / $4.25 ($0.15)NeuroLink default. Models page: "the latest version, tuned for agentic workflows (multi-step tool, browser, and long-horizon tasks) with improved coding over 1.2." and "Recommended for new work." Output figure 131072 (see below).
muse-spark-1.2Standard1,048,576yes$1.25 / $4.25 ($0.15)NeuroLink fallback. Models page: "the previous version." Home page card: "A coding-optimized model purpose-built for agentic workflows."
muse-spark-1.1Standard1,048,576yes$1.25 / $4.25 ($0.15)NeuroLink fallback. Models page: "the original version." Home page card: "Built for agentic workflows, coding, computer use & multimodal perception."
muse-spark-1.3-contributorContributor1,048,576yes$0.10 / $0.20 ($0.002)Models page: the Contributor variant "trades a lower price for permission to train on your prompts and completions". Not in models.fallbacks.
muse-spark-1.2-contributorContributor1,048,576yes$0.10 / $0.20 ($0.002)Models page: the Contributor variant "trades a lower price for permission to train on your prompts and completions". Not in models.fallbacks.

Context, tier and input modalities come from the "Available Muse Spark models" table on the Models page (retrieved 2026-09-29); prices come from the Pricing page. The Models page lists muse-image-1.0, muse-voice-transcribe-1.0 and sam-3.1 in separate per-family tables; they are not in this catalog. The Chat Completions page says "Chat Completions serves the Muse Spark text models."

Output figure: muse-spark-1.3 carries maxOutputTokens: 131072, the value in the Coding agents guide entry for that model id: "Limits: context = 1048576, output = 131072". models.defaultContextWindow (1,048,576) is the context window the Overview and Models pages state for the Muse Spark models. models.defaultMaxOutputTokens (16,384) is a NeuroLink placeholder that the vendor does not publish.

Vision: the five catalog models are vision: true. The "Available Muse Spark models" table on the Models page lists image among their input modalities, and the Image understanding page says "image understanding is fully supported" on Chat Completions.

Status: the status field of the five models holds production, a placeholder for a value the catalog schema requires; it is not a vendor statement.

Fallback order when the default is unavailable: muse-spark-1.2 → muse-spark-1.1. The runtime fallback model name the loader derives (fallbacks[1]) is muse-spark-1.1. The Contributor-tier ids are in models.catalog and models.topModels; models.fallbacks lists the Standard-tier ids.

Vendor-documented request parameters​

These are statements on Meta's own pages about request fields on Chat Completions. They are quoted as written and were not exercised against the API.

RequestWhat the vendor page says
tool_choice other than autoTool calling page: "tool_choice must be "auto": Only "auto" (the default) is supported on both Chat Completions and the Responses API; "none", "required", and named function choices return HTTP 400"
reasoning_effort: "none"Chat Completions page: "Muse Spark always reasons, so reasoning_effort: "none" returns HTTP 400."
n above 1Chat Completions page: "Only n=1 is supported; values greater than 1 return HTTP 400."
stopChat Completions page, OpenAI compatibility notes table: "Not supported on reasoning models. This matches OpenAI's behavior on o3 and o4-mini."
logprobs: trueChat Completions page: "logprobs: true returns HTTP 400"
max_tokensChat Completions page, Parameters table: "The deprecated alias max_tokens is still accepted."
system roleChat Completions page: "accepted for OpenAI compatibility and treated at the same level as developer"

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 Meta Model API — docs-verified only, not live-verified:

ProbeResult
Rosterunauthenticated GET https://api.meta.ai/v1/models answered HTTP 401 on 2026-09-29, so model ids are taken from the Models page; the roster has not been checked. The List models reference shows GET /models called with a Bearer key
Auth-failure shapeunauthenticated GET https://api.meta.ai/v1/models: HTTP 401, content-type: application/json, body {"error":{"code":"invalid_api_key","message":"Unauthorized","param":null,"type":"authentication_error"}}. An unauthenticated GET https://api.meta.ai/v1/chat/completions answered HTTP 401 with the same body and the header allow: POST
Billingpublic Pricing and Authentication and billing pages, quoted in Billing above; no signup performed, 2026-09-29
Tools / structured outputdocumented on the Tool calling and Structured output pages; neither was exercised live, and 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=meta-model-api 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​

The Vendor statement column quotes the vendor's own pages.

SymptomVendor statementNeuroLink note
HTTP 401Error handling page: "The API key was missing, invalid, or revoked." Its fix line: "Send a valid Authorization: Bearer $MODEL_API_KEY header."NeuroLink reads META_MODEL_API_API_KEY; the vendor's examples use MODEL_API_KEY
HTTP 402Error handling page: "The request cannot be completed due to a billing issue." Its fix line: "Check account status and billing details in the Model API dashboard."The example body's error.code is billing_not_configured
HTTP 404, code model_not_foundQuickstart page: "404 model_not_found: use a valid model ID such as muse-spark-1.3 (the default in these examples) or muse-spark-1.1 exactly."The five catalog ids are listed in Models
HTTP 429Error handling page: "Your team has exceeded the rate limit." Its fix line: "Implement exponential backoff with jitter."Limits are listed in Billing
HTTP 504 on a non-streaming requestChat Completions page: "a non-streaming request that runs too long returns HTTP 504". Error handling page fix line: "Stream the response by setting stream: true."Use neurolink.stream() for a streamed request

See also​