east · build
One endpoint, answered by real GPUs
TENSORRA is OpenAI wire-compatible. Swap the base URL and your requests are scheduled onto browser GPUs, verified, and settled to their operators on Solana.
curl https://api.tensorra.site/v1/chat/completions \
-H "Authorization: Bearer $TENSORRA_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "tensorra-llama-3.1-70b",
"messages": [{"role": "user", "content": "Explain proof of silicon."}]
}'models
| model | context | price | scheduled on |
|---|---|---|---|
| tensorra-llama-3.1-70b | 128k | $0.42 / Mtok | NVIDIA A100 · L40S |
| tensorra-mixtral-8x7b | 32k | $0.18 / Mtok | RTX 4090 mesh |
| tensorra-qwen2.5-7b | 32k | $0.06 / Mtok | Consumer tier |
| tensorra-embed-v2 | 8k | $0.01 / Mtok | Apple Silicon tier |
verification
Every response carries a tensorra_proof object: the run proof hash of the serving node, the verifier quorum signature, and the Solana transaction that paid the operator. Replay a sample with POST /v1/proofs/verify to confirm the FLOPs were actually executed.
endpoints
POST/v1/chat/completions
POST/v1/completions
POST/v1/embeddings
GET/v1/models
POST/v1/proofs/verify
GET/v1/network/runs
payment rails
Prepay in USDC or $TENSORRA on Solana. Usage is metered per epoch (6h); unused balance stays in your associated token account. Inference revenue flows straight into the treasury that buys tokenized silicon equities.
rate limits
Free
20 req / min
Holder (25k $TENSORRA)
600 req / min