Models and pricing

InferNear serves NEAR AI Cloud's attested TEE chat models, and only those. The live list, with current prices, is always one call away and needs no key:

curl -s https://api.infernear.dev/v1/models

Models listed on 13 Nov 2026

Prices are NEAR AI Cloud's listed prices for each model, shown here per million tokens.

Model id	Input, per 1M tokens	Output, per 1M tokens
deepseek-ai/DeepSeek-V4.1-Flash-1017	$0.30	$1.20
deepseek-ai/DeepSeek-V4.1-Flash-1017-abliterated	$0.30	$1.20
Qwen/Qwen3.6-35B-A3B-FP8	$0.17	$1.10
Qwen/Qwen3.8-27B	$0.04	$1.35
Qwen/Qwen3.8-27B-abliterated	$0.04	$1.35
Qwen/Qwen3-VL-30B-A3B-Instruct	$0.15	$0.55
z-ai/glm-5.3-flash	$0.08	$0.50

The list and the prices follow NEAR AI Cloud and can change. Treat GET /v1/models as the source of truth, and use the exact id it returns: aliases are not accepted.

Reading the price fields

Each model in the response carries two price objects:

{
  "id": "z-ai/glm-5.3-flash",
  "inputCostPerToken": {"amount": 80, "scale": 9, "currency": "USD"},
  "outputCostPerToken": {"amount": 500, "scale": 9, "currency": "USD"}
}

amount at scale: 9 is a price in nano-USD per token. Divide by 1,000 to get dollars per million tokens: 80 is $0.08 per million input tokens.

How a call is charged

Credits are counted in micro-USD (one millionth of a dollar).

1. Before the call, the relay sets aside enough of your credit to cover the largest reply the request allows (max_tokens). You need that much available, even though you will usually be charged much less. It has two parts. The reply part, at the 4,096-token cap, is roughly $0.002 to $0.006 depending on the model. The request part is added on top: in standard mode the relay assumes the worst case of one input token per byte of your request, plus 1,024, so a very large request sets aside noticeably more.
2. After the call, you are charged for the tokens the model reports using, at the prices in the list, rounded up to a whole micro-USD. The rest of what was set aside is released.
3. The charge is recorded on NEAR under the call's inference ID. See How credits work.

If a reply completes but reports no usage, the call is charged at the amount that was set aside. If the call fails and no reply is produced, nothing is charged.

Both standard and private mode serve the same models at the same prices.