Network stats

Public read-only snapshots of the live pool.

None of these endpoints need an API key. All are served from https://api.malibu.tech.

Endpoint	What it returns
`GET /v1/stats/overview`	Network totals, a 30-minute series, and up to 90 complete UTC days
`GET /v1/stats/routability`	Per-model and per-provider routability, with a summary
`GET /v1/stats/models`	The `models` section of the routability snapshot
`GET /v1/stats/providers`	The `providers` section of the routability snapshot
`GET /v1/status`	Buyer-facing up/degraded status with per-model slot availability
`GET /v1/rate-card`	The signed per-model rate card (`/v1/rate-card.sig` for the signature)

GET /v1/network-stats is a compatibility alias for the overview body. Prefer /v1/stats/overview.

Overview

GET https://api.malibu.tech/v1/stats/overview

The snapshot is regenerated about every 30 seconds. Treat it as expired after stale_after.

{
  "generated_at": "2026-09-23T04:36:15Z",
  "stale_after": "2026-09-23T04:36:45Z",
  "network": {
    "tokens_served_total": 339127635,
    "tokens_in_total": 297491445,
    "tokens_out_total": 41636190,
    "requests_total": 230793,
    "nodes_online": 7,
    "nodes_hardware_attested": 1,
    "models_serving": 2,
    "network_utilization_pct": 21,
    "gpu_cores_total": 150,
    "cpu_cores_total": 84,
    "unified_ram_gb_total": 360,
    "bandwidth_gb_per_s": 1568,
    "network_power_kw": 0.38,
    "avg_tokens_per_request": 1469,
    "capacity_estimate_sources": {"bandwidth_gb_per_s": "estimated_from_hardware_profile_or_provider_reported_summary"}
  },
  "timeseries": {
    "rpm_30m": {"bucket_seconds": 60, "points": [{"t": "2026-09-23T04:06:00Z", "value": 2}]}
  }
}

These values are a dated snapshot from 23 Sep 2026. Call the endpoint for live numbers.

Field	Meaning
`generated_at` / `stale_after`	Snapshot time and expiry (RFC 3339)
`network.nodes_online`	Providers connected at snapshot time
`network.nodes_hardware_attested`	The subset of `nodes_online` with hardware-rooted Apple managed device attestation
`network.models_serving`	Distinct models with at least one serving provider
`network.requests_total`	Cumulative request count since stats rollup began. This is not a 24-hour window.
`network.tokens_served_total`	Cumulative prompt + completion tokens
`network.tokens_in_total` / `tokens_out_total`	Cumulative prompt vs completion split
`network.network_utilization_pct`	Share of provider slots busy at snapshot time, 0–100
`network.unified_ram_gb_total`	Sum of reported unified memory
`timeseries.rpm_30m`	Requests per minute in 60-second buckets, last 30 minutes
`timeseries.tpm_30m`	Tokens per minute in 60-second buckets, last 30 minutes. Each point carries `input_tokens` and `output_tokens`
`timeseries.daily_90d`	Up to 90 complete UTC days, oldest first. The UTC day of the snapshot is omitted. `bucket` is `1d`. Each point has `t` (`YYYY-MM-DD`), `requests`, `input_tokens`, and `output_tokens`. A quiet day inside the window is zero
`idle_prewarm`	Idle pre-warm diagnostics: the share of the pool with pre-warm active, and skip counts by reason over the last hour

Capacity fields (bandwidth_gb_per_s, network_power_kw, core counts) are estimates from hardware profiles. capacity_estimate_sources names the method for each one. They are not metered telemetry. For measured throughput, use timeseries.tpm_30m.

Routability

GET https://api.malibu.tech/v1/stats/routability

This endpoint shows which models can take traffic right now, without an API key:

{
  "generated_at": "2026-09-23T04:37:15Z",
  "summary": {"state": "redundant", "models_total": 2, "providers_total": 7, "providers_routable": 5},
  "models": [
    {
      "model_id": "mlx-community/Llama-3.2-3B-Instruct-4bit",
      "state": "redundant",
      "routable_provider_count": 4,
      "slots_free": 4,
      "slots_total": 6,
      "max_context_tokens": 120000
    }
  ],
  "providers": [
    {"provider_ref": "provider_000001", "state": "degraded", "routable": false, "attestation": "self_signed", "receipt_validity": "present"}
  ]
}

Model state is redundant (two or more routable providers), operational (exactly one), degraded (none routable, at least one serving-capable), unknown, or offline. provider_ref is a per-snapshot alias. It is not a stable provider ID and cannot be used for routing. A snapshot older than 120 seconds returns 503 stats_stale with Retry-After: 30.

Status

GET https://api.malibu.tech/v1/status

Returns status (for example up), a degraded flag, coordinator health, pool counts (ready, degraded, draining, unavailable), and per-model slots_free, max_context_tokens, and availability.

Rate card

GET https://api.malibu.tech/v1/rate-card

See Models → Rate card, Pricing comparison, and Model catalog.

Models you can call (authenticated)

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

Returns only models with at least one connected provider. See Models.

Operations → Network stats: how aggregates are computed
Network status: live vs planned ledger
Reputation: how per-provider signals surface

Litepaper

Getting started

Buyers

Providers

Network & trust

Guides index

Ledger

Network stats

Network stats

Overview

Routability

Status

Rate card

Models you can call (authenticated)

​Network stats

​Overview

​Routability

​Status

​Rate card

​Models you can call (authenticated)

​Related

Network stats

Overview

Routability

Status

Rate card

Models you can call (authenticated)

Related