Pricing · updated 2026-10-07
Model usage at list price.
In every edition.
On SaaS you top up via Stripe and pay each provider's list price, with no platform fee, no BYOK surcharge and no subscription, and we cover the Stripe processing fee. Enterprise and private deployment use the same model prices, and the platform is priced per company. We make money on the software and add no token markup.
What each edition includes
Enterprise and private deployment include everything in SaaS. Model usage is at provider list price in every edition; what differs is the platform around it.
| Feature | SaaS Developers and small teams Model usage at list price, 0% platform fee, no subscription Get your API key | Enterprise Engineering teams inside a company, and the people who sign off Model usage at list price; the platform is priced per company Talk to sales | Private deployment Companies whose prompts may not leave their network License and support, priced per company Talk to sales |
|---|---|---|---|
| Price | |||
| Model usage | Provider list price, no markup | Provider list price, no markup | Provider list price, no markup |
| Platform fee | 0% platform fee, no subscription | Priced per company | License and support, priced per company |
| Your own provider keys (BYOK) | Free, 0% surcharge | Included | Every upstream runs on your keys |
| Billing | Top up via Stripe and pay as you go; unused balance refunded within 7 days of your request | Contract and invoice | Contract and invoice |
| Free trial | 10 calls, up to $1, no card | ||
| Models and routing | |||
| Model catalogue | 124 models: chat, image, speech, video and realtime | The same catalogue, plus your own upstreams | The same catalogue, plus models inside your network |
| OpenAI-compatible endpoint | |||
| Failover and retries | |||
| Prompt caching | |||
| Model access by team | |||
| Models in your own network | If the instance can reach them | ||
| Build | |||
| Prompt library | |||
| Playground | |||
| MCP tools for agents | |||
| API keys with limits | |||
| Application registry | |||
| Observe | |||
| Request logs | |||
| Traces and sessions | |||
| Usage analytics | |||
| Exports to S3 or OTLP | |||
| Usage by team and by intentThe gateway tags every call as coding, analysis, writing, lookup, leisure or other. Nobody fills out a form. | |||
| Intent detected on every call | |||
| Spend by team and intent | |||
| Policies per intent | |||
| Routing by intent (model=auto) | |||
| Your own intent model | |||
| Cost governance | |||
| Spend caps | |||
| Organization cap and alerts | |||
| Team budgets and forecast | |||
| Daily allowance per member | |||
| Approval flows | |||
| Cost centers and chargeback | |||
| Security and data | |||
| Prompt retention | Zero by default | Zero by default, or the policy you set | You set it: bodies off, redacted, or full |
| Region pinning | US, EU or APAC | US, EU or APAC | Wherever you run it |
| Guardrail rule packs | |||
| Block, mask, guide or detect | |||
| Judge model review | On a model you host | ||
| Data classification | |||
| Tool registry (MCP, skills) | |||
| Alerts | Spend and usage alerts by email | Plus security alerts: injection, malicious skills, unregistered MCP, data leaving. Slack, email, Feishu, PagerDuty | Plus security alerts: injection, malicious skills, unregistered MCP, data leaving. Slack, email, Feishu, PagerDuty |
| Requests leave only for allowed providers | |||
| Identity and audit | |||
| Sign-in | GitHub, Google, Apple or email; SSO (SAML / OIDC); 2FA you can require of everyone | SSO (SAML / OIDC) from your IdP; accounts bound to the IdP subject, never the email | SSO (SAML / OIDC) from your IdP; accounts bound to the IdP subject, never the email |
| SCIM provisioning | |||
| Roles | Admin and member, per workspace | Member, team head and admin; team heads approve for their own team | Member, team head and admin; team heads approve for their own team |
| Audit log | Workspace audit log with filters and export, kept 180 days | Tamper-evident hash chain, JSONL export, streaming to your SIEM; kept permanently | Tamper-evident hash chain, JSONL export, streaming to your SIEM; kept permanently |
| Prompt review grants | |||
| Member data page | |||
| Where it runs | |||
| Deployment | Shared SaaS | A dedicated instance, run by us | Your VPC or data center: gateway and console |
| Updates | Continuous | Applied by us | Releases you install |
| Data processing agreement | On request | On request | On request |
| Support | Included | Included in the license | |
How SaaS billing works
Pay-as-you-go
Zero platform fee
- No provider accounts needed. Requests run on the platform key, our keys to Anthropic, OpenAI, Google and others
- Top up your balance via Stripe; pay only for tokens consumed
- Zero surcharge on every request
- Cache reads at each provider's rate (about 10% of input price on Anthropic)
Your own keys (BYOK)
0% BYOK surcharge
- Add your own provider keys to the workspace vault
- Keep using your existing provider account
- No Synthorai surcharge (0%)
- Per-model allowlist, fallback toggle and audit log
$50 launch top-up
10% off for 30 days
- Top up $50 and get 10% off platform-key requests for 30 days
- 30-day discount window, balance never expires
- Stacks with normal usage
- Refunded within 7 days of your request
Volume discounts below list price are available: talk to us about volume.
Why it's cheaper than an aggregator
A $100 top-up, same models, same provider list prices across chat, image and audio. The gap is fees.
Sample prices
Per million tokens, USD. Exactly what each request draws from your balance, zero surcharge added. A sample from across the catalogue, pulled at build time.
| Provider | Model | Input ($ per 1M tokens) | Output ($ per 1M tokens) |
|---|---|---|---|
| Amazon | nova-2-sonic | $0.06 | $0.24 |
| ByteDance | Dola-Seed-2.0-mini | $0.1 | $0.4 |
| Anthropic | claude-haiku-5-5 | $0.1 | $0.5 |
| Anthropic | claude-fable-5-1 | $10.00 | $50.00 |
| OpenAI | gpt-5.4-pro | $30.00 | $180.00 |
google-tts-chirp3-hd | $30.00 | $0 |
Compare all 124 model prices, live → Filter by provider and sort by price. Prices come from the gateway's own catalogue.
How a request is billed
Top up your balance via Stripe. Requests draw it down at the provider's per-token price.
Every request follows the same path: we pre-deduct an estimate, forward the call, then settle against the usage the provider reports. If a request fails mid-stream, an in-flight ledger releases the hold automatically and refunds the difference. A crash on our side must never corrupt your balance. That rule is enforced in code and described in our gateway cache-billing audit.
Your balance never expires, and the unused part is refunded within 7 days of your request.
What “0% BYOK” means
Bring your own provider keys and requests bypass your Synthorai balance entirely. You pay Anthropic, OpenAI or Google directly at list price, and Synthorai adds nothing on top. The 0% is not a promotion.
Keys live in a per-workspace encrypted vault, and you can allowlist them per model. With the “never fall back to the platform key” flag on, requests use only your key. If that key runs out, the request fails and your balance is not charged.
BYOK is free on SaaS. What we charge for is the Enterprise and private deployment platform.
How caching changes your bill
A cache read is input the provider has already cached. On Anthropic's schedule it costs about 10% of the input price, and a cache write about 125%. Each provider sets its own rates and we pass them through unchanged. Cross-provider translation is covered in our prompt caching guide.
Caching only pays off if repeat requests reach the upstream (the provider account a request is sent to) that holds the cache. Synthorai pins cache-enabled traffic to one upstream key per workspace. Estimate your savings with the LLM cost calculator.
FAQ
Is there a subscription or monthly minimum?
No. Top up once and requests draw from the balance as you make them. It never expires.
Are there any platform fees?
None. Zero platform fee, zero BYOK surcharge, and the Stripe processing fee is on us. Your balance is debited by exactly what each request consumes.
Is there a free trial?
New accounts get 10 trial calls with up to $1 of credit, no card required. After that it is pay as you go.
Do prices differ from going to the provider directly?
No. Per-token prices mirror each provider's list price, and the catalogue is synced from provider pricing at build time.
What about refunds?
Unused balance is refundable, no questions asked. We issue the refund within 7 days of receiving your request, and Stripe processes it.
How is Enterprise priced?
Model usage stays at list price, as on SaaS. The platform, a managed dedicated instance with budgets, approvals, guardrail rules, SSO / SCIM and audit retention, is priced per company. Talk to sales for a quote.
Can we run it inside our own network?
Yes. Private deployment puts the gateway and the console in your VPC or data center, licensed per company with support included. Prompts leave only for the providers you allow, and model usage runs on your own provider keys.
Start with whatever fits
Sign up free and make 10 trial calls without a card. Top up when you need more, and use the $50 promo when you're ready.