Models

GET /v1/models — list the models connected providers are serving right now.

An OpenAI-shaped /v1/models. It returns the models that have at least one routable provider right now, so the list changes as Macs come online and go offline. Authentication required. Unauthenticated calls return 401 missing_bearer_token.

Endpoint

GET https://api.malibu.tech/v1/models
Authorization: Bearer <MALIBU_API_KEY>

Response

This is a live response from 23 Sep 2026, trimmed, after the 17-model signed catalog went live. Call the endpoint for the current set.

{
  "object": "list",
  "data": [
    {
      "id": "mlx-community/Llama-3.2-3B-Instruct-4bit",
      "object": "model",
      "provider_count": 4,
      "total_slots": 4,
      "max_context_tokens": 120000,
      "hash_verified": true,
      "hash_verification": {"status": "all_verified", "catalogued": true, "verified_provider_count": 4}
    },
    {
      "id": "mlx-community/Qwen3.6-27B-4bit",
      "object": "model",
      "provider_count": 1,
      "total_slots": 8,
      "max_context_tokens": 200000,
      "hash_verified": true,
      "hash_verification": {"status": "all_verified", "catalogued": true, "verified_provider_count": 1}
    }
  ],
  "tier1_disclosure": {
    "plaintext_to_provider": true,
    "model_identity": "provider_reported",
    "hardware_attestation": "partial",
    "model_hash_verified": "all",
    "provider_leg_encryption": "all",
    "sticky_affinity": {"enabled": true, "ttl_seconds": 1800, "description": "..."}
  },
  "tier2": {
    "attestation": {"state": "partial", "attested_provider_count": 1, "unsupported_provider_count": 4},
    "encrypted_leg": {"state": "all", "scope": "coordinator_to_provider_only"},
    "model_hash": {"state": "all", "require_verified": true}
  }
}

Field	Meaning
`id`	The model ID to send as `model`. It is the ID the serving provider reports.
`provider_count`	Connected providers advertising this model. No per-provider rows are returned.
`total_slots`	Concurrent request slots those providers advertise.
`max_context_tokens`	Context limit for routing.
`hash_verification`	Whether providers’ reported model-weight hashes match the signed catalog. `catalogued: false` means the model has no catalog hash to compare against yet.
`compute_integrity`	Present on each entry. It currently reports `status: "unavailable"`. No live compute-integrity telemetry backs it yet.
`tier1_disclosure`	The trust posture: prompts are plaintext to the provider, and model identity is provider-reported. It also shows whether sticky routing is on and its TTL.
`tier2`	Pool-wide counts for hardware attestation, coordinator-to-provider leg encryption, and model-hash verification. On 23 Sep 2026, 1 of the 5 providers counted here was hardware-attested (`partial`). These counts cover providers advertising a listed model. The public `nodes_online` stat (7 at the time) counts every connected Mac, including ones not serving a listed model. Leg encryption covers the coordinator-to-provider hop only, and the provider still reads your prompt.

Model identity is provider-reported. Hash matching is a check against the catalog. It is not hardware attestation, and it cannot detect a provider that falsifies its own hash measurement. See Security & trust model.

From Python

from openai import OpenAI

client = OpenAI(base_url="https://api.malibu.tech/v1", api_key="<key>")
for model in client.models.list().data:
    print(model.id)

Availability

A model appears when at least one routable provider is serving it. If no routable provider is left, the model drops off the list. Requests for a model no provider advertises return 404 model_not_found. Requests for a listed model whose slots are all busy wait briefly for a free slot, then return 503 no_provider_available. For the same availability without a key, use the public routability snapshot (GET /v1/stats/models).

Listed now vs catalog

/v1/models shows only what is routable right now: models with at least one routable provider. Both warm models in the response above are covered by the live 17-model signed catalog (catalogued: true). For priced rows and the signed catalog, including models no provider has loaded, see Operations → Model catalog. A public, OpenRouter-format listing of every priced catalog row is at GET https://api.malibu.tech/v1/openrouter/models (no key needed). A row there does not mean a provider is serving it.

Rate card

Per-model pricing, public:

curl -s https://api.malibu.tech/v1/rate-card | jq .

Each row has prompt_rate_per_mtok, completion_rate_per_mtok, and prompt_cache_hit_rate_per_mtok in credits per million tokens. usd_per_million_credits (currently 1.0) converts credits to USD. Rate-card keys are normalized slugs, not serve IDs: mlx-community/Llama-3.2-3B-Instruct-4bit bills against the meta-llama/llama-3.2-3b-instruct row. A model with no matching row falls back to the default row. The detached signature is at GET /v1/rate-card.sig. Dated comparison: Pricing comparison.

Litepaper

Getting started

Buyers

Providers

Network & trust

Guides index

Ledger

Models

Models

Endpoint

Response

From Python

Availability

Listed now vs catalog

Rate card

​Models

​Endpoint

​Response

​From Python

​Availability

​Listed now vs catalog

​Rate card

Models

Endpoint

Response

From Python

Availability

Listed now vs catalog

Rate card