What actually changes if you switch.
You already know OpenRouter. This page isn’t trying to talk you into considering SayGM — it’s the numbers for someone deciding whether to switch: what you’d pay on the models you already use, and what you’d give up.
Live catalog pricing, Claude / GPT / Gemini.
Every row is read from SayGM’s catalog within the last five minutes — the same numbers the per-model pages and the homepage pricing table show. Nothing below is hand-typed, so it can’t go stale the way a maintained comparison page does.
| Model | Maker list $/Mtok (In/Out) | SayGM price $/Mtok (In/Out) | Saving |
|---|---|---|---|
| claude-fable-5Anthropic | $10.00 / $50.00 | $6.36 / $31.78 | 36.4% off |
| claude-opus-5Anthropic | $5.00 / $25.00 | $3.18 / $15.89 | 36.4% off |
| claude-opus-4-8Anthropic | $5.00 / $25.00 | $3.18 / $15.89 | 36.4% off |
| claude-opus-4-7Anthropic | $5.00 / $25.00 | $3.18 / $15.89 | 36.4% off |
| claude-sonnet-4-6Anthropic | $3.00 / $10.00 | $1.91 / $6.36 | 36.4% off |
| claude-sonnet-5Anthropic | $2.00 / $10.00 | $1.27 / $6.36 | 36.4% off |
| claude-haiku-4-5Anthropic | $1.00 / $5.00 | $0.64 / $3.18 | 36.4% off |
| gpt-5.6-solOpenAI | $5.00 / $30.00 | $2.32 / $13.91 | 53.6% off |
| gpt-5.5OpenAI | $5.00 / $30.00 | $2.32 / $13.91 | 53.6% off |
| gpt-5.4OpenAI | $2.50 / $15.00 | $1.16 / $6.95 | 53.6% off |
| gpt-5.6-terraOpenAI | $2.00 / $12.00 | $0.93 / $5.56 | 53.6% off |
| o3OpenAI | $2.00 / $8.00 | $0.93 / $3.71 | 53.6% off |
| o4-miniOpenAI | $1.10 / $4.40 | $0.57 / $2.27 | 48.5% off |
| gpt-5.4-miniOpenAI | $0.75 / $4.50 | $0.35 / $2.09 | 53.6% off |
| gpt-5.6-lunaOpenAI | $0.20 / $1.20 | $0.09 / $0.56 | 53.6% off |
| gpt-5.4-nanoOpenAI | $0.20 / $1.25 | $0.09 / $0.58 | 53.6% off |
“Maker list” is the model maker’s own published rate — shown only for a model the registry’s own price_authority record confirms SayGM’s rate mirrors. “SayGM price” is the best rate a currently-serving provider is bidding, never more than list. We have not measured an OpenRouter price for these models and do not publish one here — see how OpenRouter itself describes its pricing below.
What their own docs say
OpenRouter publicly documents that it passes through each provider’s own per-token rate with no markup on inference, and instead charges a fee when you purchase credits — 5.5% (minimum $0.80) by card, 5% by crypto at the time of writing. That’s their documented model, stated neutrally; we haven’t independently measured it and it may change — check their docs for the current figures.
The route is sealed. The lab still sees your prompt.
SayGM’s gateway runs inside an Intel TDX trusted execution environment, and the attestation quote proving that is independently verifiable. For a frontier model — Claude, GPT, or Gemini — routing through SayGM adds no additional party who can read your prompt versus calling the lab directly: the operators and the host machine never see it either way. The lab itself still sees your prompt in plaintext, exactly as it would if you called it directly.
how the attestation works →What you give up.
A switcher will find these anyway, so here they are plainly.
A far smaller catalog.
OpenRouter aggregates models across dozens of providers, including a long tail of open-weights and niche models. SayGM’s catalog is narrower by design: the frontier labs’ own models, plus a curated set of open-weights models SayGM runs confidentially inside its own TEE — not the whole aggregator long tail.
No cross-provider translation — each model speaks its own native surface.
SayGM doesn’t translate one API shape into another: a buyer pointed at the wrong surface gets a clear error, never a silently-reshaped request. That means picking the right base URL per model family:
- Anthropic — Anthropic Messages, served at
/v1/messages - OpenAI — OpenAI Chat Completions, served at
/v1/chat/completions
- Anthropic — Anthropic Messages, served at
See it on your own models.
Get a key, swap the base URL, and compare the response — same request shape as the OpenAI, Anthropic, and Google SDKs you already have.