Independent comparison · 9 Oct 2026
Every other provider charges list price or more. SayGM charges less.
Five ways to buy the same Claude, GPT and Gemini tokens. OpenRouter and Requesty add 5 to 5.5 percent. Vercel and the labs themselves charge list. SayGM bills 11.1 to 30.5 percent below it, through a gateway sealed inside a hardware enclave.
All six criteriaWhere the alternatives winHow this was priced
| OpenRouter | $105.50 |
|---|---|
| Requesty | $105.00 |
| Vercel AI Gateway | $100.00 |
| Lab direct | $100.00 |
| SayGM · Claude | $88.90 |
| SayGM · GPT | $74.50 |
| SayGM · Gemini | $69.50 |
Real-time API calls to frontier models, paid by card. SayGM figures use its published discount per lab; the others use their published fees.
Six criteria, five providers, one table.
SayGM leads on price and privacy, and matches buying direct on fees and API shape. The alternatives lead on catalogue size and free usage. Each row links to the evidence below.
| Criterion | SayGM | OpenRouter | Requesty | Vercel | Direct |
|---|---|---|---|---|---|
| Price per token | 11–30.5% below list (ahead on this point) | List + 5.5% top-up fee | List + 5% | List price | List price; batch at 50% |
| Cost of adding credit | None; credited in full (ahead on this point) | 5.5% by card, 5% crypto | Charged on usage instead | Card processing fees | None (ahead on this point) |
| What backs gateway privacy | Intel TDX enclave with public attestation (ahead on this point) | Privacy policy; logging opt-in | Policy; logs on by default, 30 days | Policy; content deleted after request | No gateway; provider terms |
| API surface | Native Anthropic, OpenAI and Gemini APIs on one key (ahead on this point) | One OpenAI-compatible API | OpenAI-compatible plus Anthropic Messages | AI SDK, OpenAI- and Anthropic-compatible | Native, one key per lab (ahead on this point) |
| Models listed | About 65 | Several hundred (ahead on this point) | 600+ (ahead on this point) | Hundreds, plus media (ahead on this point) | One lab each |
| Free usage | None; prepaid | Free tier, tight limits (ahead on this point) | Free models, 200 a day (ahead on this point) | Monthly free credit (ahead on this point) | Varies by lab |
Price per token
SayGM: 11–30.5% below list (ahead on this point)
- OpenRouter
- List + 5.5% top-up fee
- Requesty
- List + 5%
- Vercel
- List price
- Direct
- List price; batch at 50%
Cost of adding credit
SayGM: None; credited in full (ahead on this point)
- OpenRouter
- 5.5% by card, 5% crypto
- Requesty
- Charged on usage instead
- Vercel
- Card processing fees
- Direct
- None (ahead on this point)
What backs gateway privacy
SayGM: Intel TDX enclave with public attestation (ahead on this point)
- OpenRouter
- Privacy policy; logging opt-in
- Requesty
- Policy; logs on by default, 30 days
- Vercel
- Policy; content deleted after request
- Direct
- No gateway; provider terms
API surface
SayGM: Native Anthropic, OpenAI and Gemini APIs on one key (ahead on this point)
- OpenRouter
- One OpenAI-compatible API
- Requesty
- OpenAI-compatible plus Anthropic Messages
- Vercel
- AI SDK, OpenAI- and Anthropic-compatible
- Direct
- Native, one key per lab (ahead on this point)
Models listed
SayGM: About 65
- OpenRouter
- Several hundred (ahead on this point)
- Requesty
- 600+ (ahead on this point)
- Vercel
- Hundreds, plus media (ahead on this point)
- Direct
- One lab each
Free usage
SayGM: None; prepaid
- OpenRouter
- Free tier, tight limits (ahead on this point)
- Requesty
- Free models, 200 a day (ahead on this point)
- Vercel
- Monthly free credit (ahead on this point)
- Direct
- Varies by lab
Ahead on that criterion. Rows where the alternatives lead stay in.
01 · Price
Below list on all three frontier labs, from the first request.
SayGM prices Claude at 11.1 percent below Anthropic's list, GPT at 25.5 percent below OpenAI's and Gemini at 30.5 percent below Google's. There is no volume tier to reach first.
The discount comes from how capacity is bought. Independent providers on Bittensor bid to serve each request, the winning bid sets the rate, and every token line is capped at the maker's list price, so the rate can fall but not rise above list.
Open-weight models go further. Kimi K3 costs 67.5 percent less than Moonshot's own price.
Where it does not hold. Batch jobs bought direct cost half of list, which beats any real-time price, SayGM's included. Vercel runs promotions below list on some models.
| Model | Bar | Vs list |
|---|---|---|
| Kimi K3Open | −67.5% | |
| GLM 5.3Open | −48.8% | |
| DeepSeek V4.1 FlashOpen | −40.0% | |
| GPT-6 Luna | −25.5% | |
| GPT-6.1 Sol | −25.5% | |
| GLM 5.3 FlashOpen | −12.3% | |
| Claude Opus 5.5 | −11.1% | |
| Claude Fable 5.1 | −11.1% | |
| Claude Sonnet 5.5 | −11.1% | |
| Claude Sonnet 5 | −11.1% |
Open marks open-weight models on SayGM's open tier, which is not sealed end to end. Source: SayGM model catalogue, 9 October 2026.
Price your own workload
Pick a model and set monthly volumes. Each provider is priced on the same uncached basis.
Monthly bill by provider
- OpenRouter, fee included$211
- Requesty, 5% included$210
- Vercel AI Gateway$200
- Anthropic direct, list price$200
- SayGM$178
- A year, vs the cheapest alternative
- $265
- 11% lower than Vercel AI Gateway
- A year, vs OpenRouter
- $397
- Fee included
Uncached prices from the October 2026 snapshot. Prompt caching lowers every provider's bill; batch jobs bought direct cost half of list. Current SayGM rates are on saygm.com (opens in a new tab).
02 · Fees
No fee to load credit. The others take theirs on the way in or on the way out.
OpenRouter passes provider prices through and charges 5.5 percent when credit is bought. Requesty adds 5 percent to model cost as it is used. Either way the fee scales with spend, not with which model runs.
At $2,000 a month, OpenRouter's fee alone comes to $1,320 a year. On SayGM the full top-up becomes balance, and that balance then buys tokens below list.
Where it does not hold. Requesty charges nothing with your own provider keys, but then you pay each lab's list price directly.
| Provider | What is charged |
|---|---|
| OpenRouter | 5.5% by card ($0.80 minimum), 5% by cryptoOn each top-up, before a token is served |
| Requesty | 5%, or 0% with your own provider keysOn model cost, as tokens are used |
| Vercel AI Gateway | No markup on tokens; credits expire after a yearCard processing on purchase |
| Lab direct | NoneBilled by each lab separately |
| SayGM | Top-ups credited in fullNothing taken on deposit |
03 · Privacy
A gateway you can check, rather than one you have to trust.
The SayGM gateway runs in an Intel TDX confidential VM. HTTPS terminates inside it, so requests are decrypted only there; SayGM's operators and the host machine sit outside the seal. On request, the hardware signs a quote of exactly what is running, with keys rooted in Intel. Anyone can verify it against a nonce of their own.
Logs hold metadata only: model, provider, token counts, price and timing. Prompts and completions stay out of every record, and settlement records carry no account, key or IP address.
Where it does not hold. Claude, GPT and Gemini still run on hardware their makers control and receive the prompt as they would from a direct call. For prompts sealed end to end, use a confidential open-weight model, whose ID ends in -tee.
“Code inside it runs isolated from everything else on the machine, including its operating system, its hypervisor and the people who operate it.”
- Your application sends the request over an encrypted connection.
- The SayGM gateway runs inside an Intel TDX enclave. Inside the enclave it decrypts the request, strips account identity, optionally swaps personal data, and routes it.
- For a confidential open-weight model, inference also runs inside the enclave, so the prompt is never readable outside it.
- For a frontier model, the prompt leaves the enclave and reaches Anthropic, OpenAI or Google in plaintext, under that provider's API terms.
- An independent verifier checks the enclave's hardware-signed attestation report.
OpenRouter
A published privacy policy. Prompt logging is opt-in.
Requesty
Policy. Logging is on by default on self-serve plans, kept encrypted in the EU for up to 30 days; it can be turned off per key.
Vercel AI Gateway
Policy. Prompt and response content is deleted once a request completes; metadata is kept. Zero retention routing is opt-in on Pro and Enterprise.
Lab direct
No gateway in the path. The lab's API terms apply; zero data retention by approval.
04 · APIs and routing
Each model on its native API, behind one key.
SayGM serves the Anthropic Messages API, OpenAI Chat Completions and Responses, and Gemini generateContent. Prompt caching, extended thinking and tool use behave as each lab documents them, because the request format never changes.
Routing balances each request on latency, reliability and price, retries on another provider when one fails, and keeps a conversation on one provider where it can so the prompt cache keeps paying out. Cascade models fall back to the next model; fusion models combine several.
Where it does not hold. Code that sends Claude or Gemini through an OpenAI-format client must move those calls to the Anthropic or Gemini SDK. Batch jobs and hosted tools stay with the maker.
from anthropic import Anthropic client = Anthropic(- api_key=ANTHROPIC_API_KEY,+ base_url=SAYGM_ANTHROPIC_BASE_URL,+ api_key=SAYGM_API_KEY, )05 · Trade-offs
Where an alternative is the better choice.
SayGM lists about 65 models and has no free tier. If either matters, these are the cases where another provider fits better. Many teams run two: SayGM as the default, another gateway for what it does not list.
OpenRouter
Long-tail models
Several hundred models against SayGM's 65. Legacy and niche models are far more likely to be listed.
Free models
A free tier with tight limits, useful for prototypes. SayGM has none.
One request format
Every model speaks the OpenAI format, so code that hops between models needs no provider-specific handling.
Requesty
Enterprise controls
SSO, role-based access and custom SLAs on its Enterprise plan. SayGM is self-serve keys only.
EU data residency
A Frankfurt gateway on every plan, plus EU-region models.
Catalogue
600+ models across 20+ providers.
Vercel AI Gateway
Spend controls
Budgets per team, project, key or member, request logs and trace export.
Breadth beyond text
Hundreds of models, including image, video, speech and embeddings.
Promotions
Vercel runs promotional prices below list on some models, which can beat SayGM on a single model.
Lab direct
Work that can wait
Batch APIs from all three labs cost 50 percent of list, below any gateway's real-time price.
Contracts in your name
Zero data retention by approval, HIPAA BAAs from Anthropic, regional processing from OpenAI.
06 · Switching
The move is a base URL and a key.
Leaving OpenRouter? SayGM is currently paying off the last OpenRouter bill of teams that switch. The terms are SayGM's own, so confirm them on its site before you count on it.
- 01
Create a key and load credit
The full top-up becomes balance. There is no subscription or minimum.
- 02
Change the base URL
Point your Anthropic, OpenAI or Gemini SDK at SayGM and use model IDs from its catalogue.
- 03
Keep a second provider
Leave your current gateway in place as a fallback for models SayGM does not list.
07 · Method
How the numbers were made.
- Basis
- Every provider is priced against the model maker's list price for real-time, uncached tokens, in US dollars per million.
- Fees
- OpenRouter's 5.5% card fee and Requesty's 5% markup are added to the token price, because that is the effective rate an invoice reflects.
- SayGM rates
- Taken from SayGM's model catalogue on 9 October 2026. They move each settlement period, so treat them as a dated snapshot.
- Open-weight models
- Compared against the maker's own price. OpenRouter figures for them use its cheapest listed provider, which is usually not confidential.
Competitor terms are from each company's pricing and privacy pages, September 2026. Current SayGM pricing is on saygm.com (opens in a new tab). Every model, input and output.
08 · Questions
What people ask before switching.
What is SayGM?
SayGM is an LLM gateway. One API key reaches Claude, GPT, Gemini and a set of open-weight models. There is no subscription and no fee on credit top-ups. Prices are capped at the model maker's list price, and the gateway runs inside an Intel TDX trusted execution environment that publishes remote attestation.
Is SayGM cheaper than OpenRouter?
On the frontier models in this index, yes, by 16 to 34 percent once OpenRouter's 5.5 percent top-up fee is included. The exception is open-weight models: the cheapest OpenRouter providers for Kimi K3 and DeepSeek V4.1 Flash cost less on input tokens. Check the current rate for the model you use.
How does SayGM compare to Vercel AI Gateway, Requesty or buying direct?
Vercel AI Gateway and the labs themselves charge list price for real-time calls; Requesty adds 5 percent unless you bring your own keys. SayGM bills 11 to 30.5 percent below list depending on the lab. Buying direct still wins for batch jobs, which cost half of list, and for contracts such as HIPAA BAAs in your own name.
Does SayGM work with Cursor, Cline and Claude Code?
Yes. SayGM exposes OpenAI-compatible, Anthropic-compatible and Gemini-compatible endpoints. Any tool that accepts a custom base URL works after changing the URL and the API key.
How many models does SayGM support?
About 65 at the time of writing: the current Claude, GPT and Gemini lines plus open-weight models such as Kimi K3, GLM 5.3 and DeepSeek V4.1. OpenRouter lists several hundred. If you depend on a niche or legacy model, confirm it is in the SayGM catalogue first.
Can SayGM see my prompts?
The gateway runs in a trusted execution environment sealed from operators and host machines. Confidential open-weight models (IDs ending in -tee) also run inside the enclave, so prompts are decrypted only there. Frontier providers still receive the prompt, as with a direct API call. SayGM strips account identity and can replace personal data before sending.
Is there a free tier or minimum spend?
No free tier. SayGM is pay as you go: load credits, spend them per token. There is no fee on the top-up itself.
What is Bittensor subnet 28?
SayGM runs on Bittensor subnet 28. Capacity providers bid to serve inference, and that bidding sets the below-list price. Protocol data is public on Taostats.