§1

ERNIE-4.5-21B-A3B-Thinking

Name: ERNIE-4.5-21B-A3B-Thinking
Brand: Chinese AI labs

Small Apache-2.0 reasoning model that runs on a single modern GPU; not a line item in Qianfan's price table, so treat it as self-host-only.

Data checked 2026-09-06

Chinese AI labs ga open weights

Context window: 131.1K tokens in
Max output: —
Input: — per 1M tokens
Output: — per 1M tokens
Parameters: 21B total / 3B active MoE (64 text experts, 6 active + 2 shared, 28 layers)
Licence: Apache-2.0

A What it is for

reasoning long context

Input modalities	text
Output modalities	text

Model	Context	Max out	In $/M	Out $/M	Blended	Status
Kimi K3 `kimi-k3`	1M	—	$3.00	$15.00	$6.00	ga open
Kimi K2.7 Code `kimi-k2.7-code`	262.1K	—	$0.95	$4.00	$1.71	ga open
Kimi K2.7 Code Highspeed `kimi-k2.7-code-highspeed`	262.1K	—	$1.90	$8.00	$3.42	ga open
Kimi K2.6 `kimi-k2.6`	262.1K	—	$0.95	$4.00	$1.71	ga open
GLM-5.3 `glm-5.3`	1M	131.1K	$1.40	$4.40	$2.15	ga open
GLM-5.3-Flash `glm-5.3-flash`	1M	131.1K	$0.07	$0.25	$0.12	ga open
GLM-5.2 `glm-5.2`	1M	131.1K	$1.40	$4.40	$2.15	ga open
GLM-5 `glm-5`	200K	131.1K	$1.00	$3.20	$1.55	ga open
GLM-4.7 `glm-4.7`	200K	131.1K	$0.60	$2.20	$1.00	ga open
GLM-4.7-FlashX `glm-4.7-flashx`	200K	131.1K	$0.07	$0.40	$0.15	ga
GLM-4.7-Flash `glm-4.7-flash`	200K	131.1K	$0.00	$0.00	$0.00	ga
GLM-4.6V `glm-4.6v`	128K	—	$0.30	$0.90	$0.45	ga

Model pricing changes without notice and this page is a snapshot. Confirm against the vendor's own page before you build a budget on it.