Offers8 min readPublished: 2026-08-21Updated: 2026-08-21

SiliconFlow API Free Tier 2026: How Free Models, Limits and Credits Work

SiliconFlow does not use a classic all-models free quota. In 2026 the free tier means selected models billed at zero, fixed rate limits, and identity verification. This guide explains how to start, what is free, and when to switch to paid.

By Heizi· Founder & Editor· Published: 2026-08-21Hands-on tested

What the SiliconFlow free tier actually is

If you searched for siliconflow api free tier 2026, you are looking for ongoing zero-cost API access — not a one-time coupon. SiliconFlow's official rate-limit docs describe that access as free models: calls are billed at zero, they appear as ¥0 on the bill, and they stay free only while the live pricing page still marks them free.

That is different from Google Gemini's free tier, which gives rate-limited access to flagship models. On SiliconFlow, flagship chat models such as DeepSeek-V4-Pro remain paid. The free tier is a rotating set of smaller or specialized models, plus some embedding, image, and speech endpoints.

For the rest of the platform — registration, paid catalog, and BYOK — start from the complete SiliconFlow API key guide. This article only answers how the 2026 free tier works.

Three things people mix up: free models, new-user credit, paid Pro/

Free models are ongoing. If the pricing page shows 免费 / Free, input and output for that model ID cost nothing inside the fixed free-model rate limit. Identity verification is required to use all free models.

New-user credit is separate. SiliconFlow's June 2026 announcement states ¥16 on the China site or $1 on the global site. That balance can test paid models; it is not the free tier itself, and campaign rules can change.

Some models have a free ID and a paid Pro/ ID. On the August 21, 2026 pricing page, BAAI/bge-m3 is free while Pro/BAAI/bge-m3 is paid at ¥0.07 per million tokens. Using the wrong ID is how people think the free tier disappeared.

How to start the free tier

Create an account on siliconflow.cn (China) or siliconflow.com (global). Complete identity verification, then create a key at https://cloud.siliconflow.cn/me/account/ak. Screen-by-screen steps are in the SiliconFlow API key tutorial.

You do not need a credit card to call models currently marked free. Use the OpenAI-compatible base URL https://api.siliconflow.cn/v1 in China, or https://api.siliconflow.com/v1 outside China. Both accept the same key format.

Register and verify identity — official docs require this for all free models
Create an API key in the console
Pick a model ID still marked free on https://siliconflow.cn/pricing
Call it with the OpenAI SDK and the SiliconFlow base URL

Free models checked on August 21, 2026

The list below is taken from SiliconFlow's official pricing page on August 21, 2026. Only rows marked 免费 are included. Treat it as a snapshot: the catalog changes. DeepSeek had 6 more models collapsed; Qwen had 30 more. Those hidden rows are not listed here.

Paid on the same page and therefore not free-tier: DeepSeek-V4-Pro, DeepSeek-V4-Flash, Kimi K2.7/K2.6, MiniMax-M2.5, visible Qwen 3.5/3.6 chat models, and both Wan video models. The matching ID list also lives in the SiliconFlow free models snapshot.

Chat: THUDM/GLM-Z1-9B-0414 — input and output free
Chat: THUDM/GLM-4-9B-0414 — input and output free
Translation: tencent/Hunyuan-MT-7B — input and output free
OCR / vision-language: PaddlePaddle/PaddleOCR-VL-1.5 — input and output free
Embedding: BAAI/bge-m3 — free (Pro/BAAI/bge-m3 is paid)
Rerank: BAAI/bge-reranker-v2-m3 — free (Pro/BAAI/bge-reranker-v2-m3 is paid)
Embedding: BAAI/bge-large-zh-v1.5 — free
Embedding: BAAI/bge-large-en-v1.5 — free
Image: Kwai-Kolors/Kolors — free per image
Speech: Qwen/Qwen3-ASR-1.7B — free
Speech: TeleAI/TeleSpeechASR — free
Speech: FunAudioLLM/SenseVoiceSmall — free
Video: none of the listed Wan models were free

Free-tier limits: fixed, per model, 429 if you exceed them

Official docs say free-model rate limits are fixed. Spending more on paid models (moving from L0 toward L5) does not raise free-model RPM/TPM. Paid models scale with monthly spend; free models do not.

Limits are per account and per model, not per API key. Hitting the cap on one free model does not block a different model. Exact RPM/TPM values are in the console model square — do not copy a static number from a blog into production.

Over the cap you get HTTP 429 with a rate-limiting message. Wait, back off, or switch to a paid ID (including a Pro/ variant). Putting the same free key into Cherry Studio or other BYOK tools still uses these same account limits.

A first free-tier API call

Use a model ID that is still marked free on the live pricing page. The example below uses THUDM/GLM-Z1-9B-0414, which was listed as free on August 21, 2026. If that ID has moved to paid, swap it before you rely on the snippet.

python

from openai import OpenAI
import os

client = OpenAI(
    api_key=os.environ["SILICONFLOW_API_KEY"],
    base_url="https://api.siliconflow.cn/v1",
)

response = client.chat.completions.create(
    model="THUDM/GLM-Z1-9B-0414",
    messages=[{"role": "user", "content": "Reply with: FREE_TIER_OK"}],
)
print(response.choices[0].message.content)

OpenAI-compatible call to a SiliconFlow free-tier chat model

When the free tier is not enough

Stay on free models for prototyping, OCR trials, embeddings, and light BYOK chat. Leave the free tier when you need DeepSeek V4, higher concurrency, or a limit that can grow with spend.

Confirm the ID is still free, keep a paid fallback, and recheck https://siliconflow.cn/pricing before a launch. SiliconFlow can change which models are free; this page is a snapshot, not a contract.

FAQ

Which SiliconFlow models are free right now?

On the August 21, 2026 pricing snapshot: GLM-Z1-9B-0414, GLM-4-9B-0414, Hunyuan-MT-7B, PaddleOCR-VL-1.5, four BAAI embedding/rerank IDs, Kolors, Qwen3-ASR-1.7B, TeleSpeechASR, and SenseVoiceSmall. Recheck the live pricing page before you depend on any ID.

Does SiliconFlow have an API free tier in 2026?

Yes, but it is model-based. Selected models on the live pricing page are billed at zero inside fixed rate limits. It is not unlimited access to every SiliconFlow model.

Is SiliconFlow API completely free?

No. Free models cost ¥0; paid models (including DeepSeek V4 and most Pro/ IDs) follow the pricing page. New-user ¥16 / $1 credit is a separate campaign.