​
Pricing comparison

Methodology for the buyer savings claim in the Litepaper. Last reviewed: July 2026.
The Litepaper targets 20–40% lower cost vs hyperscaler pricing on comparable open-source workloads. This page explains how to verify that claim — and where it does not apply.

​
Live Malibu pricing

Malibu bills in credits; credits convert to USDC at payout/settlement time. The live rate card is authoritative:
Default rates (when no model-specific override applies): *Assumes usd_per_million_credits = 1 (default in rate card). Confirm on the live endpoint — the operator may change this field. Provider share (90%) affects provider payout, not buyer price. Buyers pay gross credits before the split. Full formula: Metering & billing.

​
How to run a fair comparison

  1. Pick a fixed workload — same model family, similar context length, same prompt/completion token mix. Malibu serves OSS weights; compare against OSS or equivalent-tier closed API pricing, not flagship-only SKUs.
  2. Snapshot Malibu rates — download /v1/rate-card and record effective_at or capture date.
  3. Snapshot anchor pricing — record hyperscaler or aggregator list price with date and region. Malibu does not maintain a third-party price feed; you own the anchor snapshot.
  4. Include cache effects — sticky conversations report cached_prompt_tokens at the lower cache-hit rate. Hyperscaler prompt-cache discounts differ; compare with and without prefix reuse if your agent replays context.
  5. Exclude non-comparable SKUs — frontier closed models without OSS equivalents are out of scope for “same workload” claims.

​
Worked example (illustrative)

Assume a workload of 1M prompt + 500K completion tokens at default Malibu rates and usd_per_million_credits = 1:
Compare $1.00 to your anchor’s all-in price for the same token mix. A 25% savings vs a $1.33 anchor falls inside the Litepaper’s 20–40% band.
This example uses default rates only. Model-specific overrides, global multipliers, and live usd_per_million_credits may differ. Always use the live rate card for quotes.

​
When the 20–40 percent claim may not hold

  • Low-volume or bursty traffic — minimum payouts and quota mechanics don’t affect buyers the same way, but your effective $/token can shift at small scale.
  • Frontier models without warm pool hardware — catalog entries for very large models (e.g. 120B+) require matching provider RAM; if the pool is cold, compare availability separately from price.
  • Latency-sensitive workloads — consumer Mac routing may trade cost for tail latency vs dedicated GPU clusters.
  • Private / compliance requirements — Malibu is cooperative-trust, not confidential compute. The comparison is economic, not security-equivalent.

​
Malibu-maintained snapshot policy

Malibu will publish dated rate-card snapshots on this page when a formal comparison table is added. Until then:
  • Buyers: use live /v1/rate-card + your anchor snapshot.
  • Docs reviewers: the Litepaper links here instead of asserting a fixed savings percentage without evidence.