Tokens per dollar

Qwen3-Coder-30B on one RTX 5090. Measured, priced, compared.

9,024

output tokens/second, one card, 320 requests at once

$0.0675 / $0.27

per 1M input / output tokens

3.7M

output tokens per dollar

822 ms

median time to first token at full load

Input tokens per dollar

Synap14.8M

FlexAI14.3M

Novita14.3M

SiliconFlow14.3M

AWS Bedrock6.7M

Alibaba3.4M

Provider	Input /1M	Output /1M	Source
Synap	$0.0675	$0.27	live price list
FlexAI	$0.070	$0.260	flex.ai
Novita	$0.070	$0.27	novita.ai
SiliconFlow	$0.070	$0.28	OpenRouter
AWS Bedrock	$0.15	$0.60	OpenRouter
Alibaba	$0.2925	$1.46	OpenRouter

We are lowest on input and 4% above the lowest on output. Both legs are shown because both are billed; the API's own /v1/models is the authority on what you pay.

Measured 2026-10-03, 5,120 requests at a concurrency of 320, zero errors. Prices checked 2026-10-08. The 9,024 figure is the card's total output while it serves many requests together — it is not the speed of any one request. One request on its own ran at about 29 tokens/second in that run, and 86 tokens/second re-measured against the live endpoint on 2026-10-07.

Try it

curl https://pool.linkrra.com/v1/chat/completions \
  -H "Authorization: Bearer $LINKRRA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3-coder-30b",
    "messages": [{"role": "user", "content": "Write a binary search in Python."}]
  }'