Find a page

Where do you want to go?

The whole lineup, live.

Browse the live catalog and switch models without changing your endpoint or API key. Each row shows checkpoint quantization and a guaranteed token-price ceiling; completed-request accounting can only charge less.

Coding models

/
8 / 8
QuantizationMax outputCache ceiling / 1M

kimi-k3

Multimodal 2.8T MoE with 1M context. The plan's flagship reasoning model.

MXFP4 (native)1,048,57665,536$2.415$0.2415$12.075

glm-5.3

Reasoning is always on; reasoning_effort selects how much. 1M-token context.

FP8 (native)1,048,57665,536$1.16725$0.116725$3.6225

qwen-3.8-27b

Dense 27B VL model with speculative decoding. The best value per dollar in the catalog.

FP8262,14465,536$0.36225$0.20125$2.576

deepseek-v4-pro

High-capacity reasoning model with 1M-token context.

FP4+FP8 mixed1,048,576131,072$0.805$0.0805$2.415

deepseek-v4.1-flash

Fast reasoning model with vision and 1M-token context.

FP4+FP8 mixed1,048,576131,072$0.12075$0.0161$0.483

glm-5.3-flash

Vision-capable fast tier of GLM 5.3 with 1M-token context.

FP8 (native)1,048,57665,536$0.12075$0.02415$0.4025

gemma-4-31b

Multimodal (text + vision) with tool calling. The highest request volume per allowance dollar.

NVFP4262,14465,536$0.1127$0.00805$0.3381

deepseek-v4-flash

Speculative decoding and 1M-token context.

FP4+FP8 mixed1,048,576131,072$0.1127$0.02254$0.2254

Every model combined$4.03 / 7d · $8.05 / 30d(Lite plan — one shared pool)

Completed requests use measured request cost. Token figures are USD-per-million safety ceilings: a completed request with accounting metadata costs the same or less. Quantization names the checkpoint precision, so competitor comparisons must match it.

Z-Image-Turbo

Image generation, included in every plan.

$0.005/ image

Uses your plan allowance. No additional charge.

Precision
BF16
Default steps
9
Supported sizes
1024 × 1024, 896 × 1152, 768 × 1152, 1152 × 896, 1216 × 704

5 free images each day, resetting at midnight KST. After the free set, both your plan's image cap and its shared text-and-image allowance apply. Every supported resolution has the same per-image price.

Create imagesView plansPOST /v1/images/generations