Frontier models,
at or below list.
Vercel AI Gateway passes through the provider’s list price with no markup. On SayGM, providers compete to serve each request, so Claude, GPT and Gemini bill at or below list, through a gateway sealed inside a Trusted Execution Environment.
At or below list price,
by design.
Vercel charges what the provider charges. SayGM charges at or below the model maker’s published list price, at any volume and from your first request, because independent providers compete to serve each request and that competition sets the rate.
Funding your balance costs the amount you fund it with. You pay the rate of the provider that serves each request, never above list.
A sealed gateway is something you can check.
Vercel publishes a clear policy: it deletes prompt and response content once a request completes. SayGM’s gateway runs inside an Intel TDX trusted execution environment, and the attestation quote proving it is independently verifiable, so your prompt stays out of reach of its operators and the host machine.
For Claude, GPT and Gemini, the provider serving the model still sees your prompt, exactly as it would if you called it directly, under its own API terms. Confidential models run sealed end to end.
Verifiable by anyone
Remote attestation lets you confirm for yourself exactly what code the gateway is running.Sealed from the routing layer too
Most privacy claims stop at the model. This one covers the gateway itself.Built for the audit trail
Pairs with signed logs of which model handled what, for teams that need to show their work.What you could save.
Pick a model, set your monthly volume, see the delta.Most input tokens in real traffic are cache reads, and both gateways price them well below a fresh one — so the share matters more to the bill than the totals do. The default is what SayGM measures across its own traffic.
Price per million tokens.
USD per 1M tokens, uncached input / output. SayGM’s side is the best rate a provider offered, as of 52 minutes ago.Vercel’s is the per-model rate its public model list publishes, as of 27 minutes ago, with no fee added because Vercel charges none.SayGM vs. Vercel AI Gateway, line by line.
Common questions about switching.
Does Vercel AI Gateway mark up token prices?
No. As of September 2026 Vercel charges the provider’s list price with no markup and no platform fee on tokens, on the free and paid tiers and with your own keys, as its pricing docs set out. It also runs promotional prices below list on some models and negotiates volume discounts. SayGM prices Claude, GPT and Gemini at or below the model maker’s list price, because providers compete to serve each request and that competition sets the rate.
Is SayGM cheaper than Vercel AI Gateway?
Across the frontier models this page compares, 15% for the middle one, against the rate Vercel’s public model list publishes. It varies by model, and the calculator above prices your exact volumes. While a Vercel promo runs, a single model can cost less there, and the table shows that as it is.
Does Vercel AI Gateway have a free tier?
Yes. As of September 2026 every Vercel team gets a monthly free credit that covers a subset of models at lower per-model rate limits, and it stops applying once you buy credits, per Vercel’s FAQ. For small experiments that credit can be worth more than any per-token saving. SayGM is pay as you go from a prepaid balance, from the first request.
Does Vercel AI Gateway have rate limits?
On the free tier, yes: lower limits per model. On the paid tier Vercel applies none of its own and the upstream provider’s limits still apply, as of September 2026 per its rate limit docs. SayGM sets limits per key, and you can ask for a higher one.
Can I bring my own provider keys?
On Vercel, yes. BYOK works on the paid tier with no fee, and a request that fails on your key is retried on Vercel’s credentials and charged to your credits. SayGM supplies capacity from its own providers. If you hold a negotiated rate with a lab that beats SayGM’s price on a model, keeping that model on your own key is the cheaper choice.
Does Vercel AI Gateway store my prompts?
Vercel says it does not train on prompts and deletes prompt and response content once the request completes, keeping request metadata, per its FAQ. Routing only to providers that agree to zero data retention is an opt-in on Pro and Enterprise. SayGM works at a different layer: the gateway runs inside an Intel TDX trusted execution environment with a verifiable attestation quote, so its operators and host machines cannot read your prompt. Claude, GPT and Gemini providers process prompts under their own API terms, and confidential models run sealed end to end.
Can I use SayGM from the Vercel AI SDK?
Yes for GPT models: the AI SDK’s OpenAI provider takes SayGM’s base URL and your SayGM key, and standard text requests work as before. Claude and Gemini are served on their own native APIs, so call them with a client for those. Vercel’s gateway supports the AI SDK directly along with OpenAI- and Anthropic-compatible APIs. You can keep it for breadth while sending frontier traffic to SayGM.
Where is Vercel AI Gateway the better choice?
For breadth, spend controls and platform fit. It carries hundreds of models across text, image, video and audio, sets budgets per team, project, key or member, and signs in with OIDC on Vercel deployments, while working from any other host too. SayGM is built for teams that want frontier models at or below list through a gateway whose privacy they can verify.