Pricing

Plans for indie hackers, AI-native startups, and enterprises. Pay only for what you use — no minimum spend.

Compare plans

Feature Free Pay-as-you-go Regulated EU
Pricing & usage
Monthly cost$0$0 + usageCustom contract
Pricing modelFreeToken cost + 5.5%Negotiated rate card
Monthly requests1,000UnlimitedUnlimited + committed throughput
Spend caps & alerts
Models & routing
Model catalogOpen-weight only100+ across every major providerFull catalog + private endpoints
Smart retries & fallback chains
Prompt & semantic caching
Bring Your Own Key (BYOK)KMS-wrapped + HSM
Compliance & governance
Residency profilesGlobal onlyEU, FR, DE, UKCustom (HDS, SecNumCloud, BSI C5)
Zero Data RetentionWhere supportedEnforced + attestation
Risk classification (Art. 5 / Annex III)
Signed audit vault (WORM)30-day retention10-year retention + Merkle anchoring
Signed deployer policy engine
Human oversight queues (Art. 14)
Auto-generated FRIA (Art. 27)
Article 73 incident reporting
Regulator access keys
Security & identity
SOC 2 Type II
SSO (SAML / OIDC)
SCIM 2.0 provisioning
Support
Community (Discord, GitHub)
Email support✓ (priority)
Compliance success manager
Uptime SLA99.9%99.95%

Wondering about exact token costs? On Pay-as-you-go we pass through the base costs of every underlying model completely transparently, plus a flat 5.5% routing fee. No hidden markups.

View all supported models and token pricing

Frequently asked questions

How are tokens billed?

Input and output tokens are billed per model at the posted provider rates. The token counts and per-model prices are surfaced on every response in the compliance_metadata envelope so your invoicing always matches what your application observed.

Do you mark up provider pricing?

No. We pass through the published provider rate card unchanged. The price you see in the model catalog is exactly what you pay for tokens — the same as on the provider's own website.

Are failed or fallback attempts billed?

No. When the smart retries or fallback chain kicks in, you're billed only for the successful model run. Failed attempts and re-validations are absorbed by the platform.

What is "Regulated EU" and who needs it?

Regulated EU is the tier built for organizations operating high-risk AI systems under the EU AI Act, GDPR, DORA, MiFID II, MDR, or HDS. It adds the signed deployer policy engine, the WORM audit vault with regulator access keys, auto-generated FRIA and Article 73 incident drafts, human oversight queues, and a named compliance success manager.

Is there a minimum spend or lock-in on Pay-as-you-go?

No minimums and no lock-in. You pay only for what you use. Spend caps, alerts, and webhooks are included by default so you stay in control.

Do you train on customer data?

Never. Meridian Blue does not train on your data. Upstream provider retention is enforced via the Zero Data Retention layer where supported — and on Regulated EU, ZDR is contractually attested per provider.

Can I separate environments (dev, staging, production)?

Yes. Create separate API keys per environment with their own scopes, spend caps, residency profiles, and risk-tiered rate limits. Test keys (mb_test_sk_) only route to free preview models so you can wire up CI safely.

What happens if a provider is down or a model errors?

Smart retries handle transient failures inside the gateway. If they exhaust, the fallback chain promotes to the next eligible candidate — re-classified and re-validated against your residency and ZDR requirements before being used. You're billed only for the successful run.

What payment methods do you accept?

Pay-as-you-go accepts credit and debit cards and SEPA direct debit. Regulated EU supports invoicing, purchase orders, and annual contracts in EUR.