> ## Documentation Index
> Fetch the complete documentation index at: https://docs.siliconflow.cn/llms.txt
> Use this file to discover all available pages before exploring further.
# Release notes
### \[Model Service Adjustment] GLM-4.7, Kimi-K2.5, and Other Models to Be Discontinued
To further optimize resource allocation and provide more advanced and high-quality technical services, the platform will discontinue the following models on 2026-06-11:
* Pro/moonshotai/Kimi-K2.5 (requests will be redirected to K2.6)
* Pro/zai-org/GLM-5 (requests will be redirected to GLM-5.1)
* Pro/zai-org/GLM-4.7
* netease-youdao/bce-embedding-base\_v1
* netease-youdao/bce-reranker-base\_v1
After `Pro/moonshotai/Kimi-K2.5` and `Pro/zai-org/GLM-5` are discontinued, requests will be redirected to `Pro/moonshotai/Kimi-K2.6` and `Pro/zai-org/GLM-5.1` respectively, and billed according to the corresponding model metering standards.
If you are currently using any of the above models, it is recommended that you switch to alternative models as soon as possible to avoid service disruptions.
### Platform Service Adjustment Notice
To further optimize resource allocation and provide more advanced and high-quality technical services, the platform will discontinue the following models on 2026-05-15:
* moonshotai/Kimi-K2-Thinking
* moonshotai/Kimi-K2-Instruct-0905
* Pro/moonshotai/Kimi-K2-Instruct-0905
* Pro/moonshotai/Kimi-K2-Thinking
* zai-org/GLM-4.6
* zai-org/GLM-4.6V
* THUDM/GLM-Z1-32B-0414
* THUDM/GLM-4.1V-9B-Thinking
* inclusionAI/Ring-flash-2.0
* Qwen/Qwen3-30B-A3B-Thinking-2507
* Qwen/Qwen3-235B-A22B-Instruct-2507
If you are currently using any of the above models, it is recommended that you switch to alternative models as soon as possible to avoid service disruptions.
### Platform Service Adjustment Notice
* To further protect the security of your account and platform rights, and to enhance service safety and reliability, accounts that have not completed real-name verification will be unable to use platform functions starting from May 15, 2026. The restriction will be lifted only after verification is completed.
### Platform Service Adjustment Notice
To further optimize resource allocation and provide more advanced and high-quality technical services, the platform will discontinue the following models on 2026-04-29:
* Kwaipilot/KAT-Dev
* PaddlePaddle/PaddleOCR-VL
* Qwen/QwQ-32B
* Qwen/Qwen2.5-VL-32B-Instruct
* Qwen/Qwen2.5-VL-72B-Instruct
* deepseek-ai/DeepSeek-R1-Distill-Qwen-32B
* deepseek-ai/DeepSeek-R1-Distill-Qwen-14B
* deepseek-ai/DeepSeek-R1-Distill-Qwen-7B
* Qwen/Qwen2.5-Coder-32B-Instruct
* Qwen/Qwen2-VL-72B-Instruct
* internlm/internlm2\_5-7b-chat
* IndexTeam/IndexTTS-2
If you are currently using any of the above models, it is recommended that you switch to alternative models as soon as possible to avoid service disruptions.
### Platform Service Adjustment Notice
To further optimize resource allocation and provide more advanced and high-quality technical services, the platform will discontinue the following models on 2026-04-22:
* Qwen/Qwen3-Coder-480B-A35B-Instruct
* Qwen/Qwen3-235B-A22B-Thinking-2507
* Qwen/Qwen3-VL-235B-A22B-Thinking
* Qwen/Qwen3-VL-235B-A22B-Instruct
* deepseek-ai/DeepSeek-V2.5
* baidu/ERNIE-4.5-300B-A47B
* ascend-tribe/pangu-pro-moe
If you are currently using any of the above models, it is recommended that you switch to alternative models as soon as possible to avoid service disruptions.
### Platform Service Adjustment Notice
To further optimize resource allocation and provide more advanced and high-quality technical services, the platform will discontinue the following models on 2026-03-17:
* Pro/MiniMaxAI/MiniMax-M2.1
* Pro/Qwen/Qwen2-7B-Instruct
* Qwen/Qwen2-7B-Instruct
* Pro/THUDM/glm-4-9b-chat
* THUDM/glm-4-9b-chat
* deepseek-ai/deepseek-vl2
* Pro/Qwen/Qwen2.5-VL-7B-Instruct
* Qwen/Qwen3-Next-80B-A3B-Thinking
* Qwen/Qwen3-Next-80B-A3B-Instruct
* Qwen/Qwen2.5-Coder-7B-Instruct
* Pro/Qwen/Qwen2.5-Coder-7B-Instruct
If you are currently using any of the above models, it is recommended that you switch to alternative models as soon as possible to avoid service disruptions.
### Platform Service Adjustment Notice
To ensure the rationality and consistency of the platform's model pricing system, the platform has adjusted the pricing for the Qwen/Qwen3.5-397B-A17B model. For details, please visit ["Model Square"](https://cloud.siliconflow.cn/me/models).
Thank you for your support and understanding!
### Platform Service Adjustment Notice
To further optimize resource allocation and provide more advanced and high-quality technical services, the platform will discontinue the following models on 2026-02-09:
* MiniMaxAI/MiniMax-M2
* MiniMaxAI/MiniMax-M1-80k
* moonshotai/Kimi-Dev-72B
* Pro/THUDM/GLM-4.1V-9B-Thinking
* Tongyi-Zhiwen/QwenLong-L1-32B
* Qwen/QVQ-72B-Preview
* THUDM/GLM-Z1-Rumination-32B-0414
* Pro/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B
* Qwen/Qwen3-30B-A3B
* stepfun-ai/step3
If you are currently using any of the above models, it is recommended that you switch to alternative models as soon as possible to avoid service disruptions.
### Platform Service Adjustment Notice
To further optimize resource allocation and provide more advanced and high-quality technical services, the platform will discontinue the following models on 2025-12-31:
* zai-org/GLM-4.5
* Qwen/Qwen3-235B-A22B
If you are currently using any of the above models, it is recommended that you switch to alternative models as soon as possible to avoid service disruptions.
### Notice on Service Adjustment for Platform Credit Balance
To better protect your platform rights and improve resource usage efficiency, the platform has made the following adjustments to the display format of the **Credit Balance** service:
* **Credit balance used before November 30, 2025** will be uniformly converted into a used-up voucher:
* The **total amount and used amount** of this voucher are equal to your historically accumulated consumed credit balance;
* The **remaining available amount of this voucher is currently 0**. It is solely for recording historical entitlements and will not affect subsequent transactions.
* **Credit balance obtained before November 30, 2025, but not yet used** will be converted into an available voucher:
* The total amount of this voucher is consistent with the remaining credit balance as of November 30;
* Currently, the applicable scope of this voucher remains the same as the previous credit balance and can be used normally for deductions. Any future adjustments to the applicable scope will be based on the voucher's description; - This voucher is valid until 23:59:59 on December 31, 2099.
* **After November 30, 2025, platform incentives will be issued in the form of vouchers.**
You can go to 【Balance Top-up > Vouchers】, click on the voucher count to view the voucher list and details.
### Platform Service Adjustment Notice
To further optimize the model service quality, the platform will gradually update the `Deepseek-V3.2-Exp` model to the `Deepseek-V3.2` version over the next two days. Requests to `Pro/deepseek-ai/DeepSeek-V3.2-Exp` and `deepseek-ai/DeepSeek-V3.2-Exp` will be redirected to `Pro/deepseek-ai/DeepSeek-V3.2` and `deepseek-ai/DeepSeek-V3.2`, respectively.
### Platform Service Adjustment Notice
To further optimize resource allocation and provide more advanced and high-quality technical services, the platform will discontinue the following models on `2025-11-20`:
* inclusionAI/Ling-1T
* inclusionAI/Ring-1T
If you are currently using any of the above models, it is recommended that you switch to an alternative model as soon as possible to avoid any impact on your services.
### Platform Service Adjustment Notice
To further optimize resource allocation and provide more efficient and stable computing power services,
the platform will adjust the Rate Limits for certain models starting from `November 11, 2025`.
The models affected by this adjustment are: `Pro/deepseek-ai/DeepSeek-R1`, `Pro/deepseek-ai/DeepSeek-V3`, `Pro/deepseek-ai/DeepSeek-V3.1-Terminus`, `zai-org/GLM-4.6`,`inclusionAI/Ling-1T`，`inclusionAI/Ring-1T`,`MiniMaxAI/MiniMax-M2`;
If your business has specific requirements for high concurrency or large-scale throughput, please [contact us](https://siliconflow.feishu.cn/share/base/form/shrcnxIMbsUUDf7xjrRIjenRYoh?hide_user_id=1\&prefill_user_id=sfd17aubjk20jc73ada8ng) to apply for a higher quota.
Thank you for your understanding and support.
### Platform Service Adjustment Notice
To further optimize resource allocation and provide more efficient and stable computing services, the platform will close the purchase entry for Usage Tiers starting from November 7, 2025.
This adjustment only affects the availability of new purchase entries. Your existing Usage Tiers, current usage level, and the platform's automatic upgrade/downgrade mechanism based on consumption amount will remain unaffected.
If you have a need to quickly increase your usage tier or raise your Rate Limits, please [contact us](https://siliconflow.feishu.cn/share/base/form/shrcnxIMbsUUDf7xjrRIjenRYoh?hide_user_id=1\&prefill_user_id=sfd17aubjk20jc73ada8ng).
Thank you for your understanding and support.
### Platform Service Adjustment Notice
To further optimize resource allocation and provide more advanced and high-quality technical services, the platform will discontinue the following models on 2025-10-09:
* deepseek-ai/DeepSeek-V3.1
* Pro/deepseek-ai/DeepSeek-V3.1
If you are using any of the above models, it is recommended that you switch to V3.1 Terminus as soon as possible to avoid any disruption in service.
### Platform Service Adjustment Notice
To further optimize the quality of model services, the platform updated the moonshotai/Kimi-K2-Instruct and Pro/moonshotai/Kimi-K2-Instruct models to the latest 0905 version on September 15. The previous 0711 version will no longer be available.
The moonshotai/Kimi-K2-Instruct and Pro/moonshotai/Kimi-K2-Instruct models have been removed from the Model Plaza. All corresponding model requests will now be directed to moonshotai/Kimi-K2-Instruct-0905 and Pro/moonshotai/Kimi-K2-Instruct-0905, respectively.
### Platform Service Adjustment Notice
To further optimize resource allocation and provide more advanced and high-quality technical services, the platform will take the following models offline on September 4, 2025:
* tencent/HunyuanVideo-HD
* Wan-AI/Wan2.1-I2V-14B-720P-Turbo
* Wan-AI/Wan2.1-I2V-14B-720P
* Wan-AI/Wan2.1-T2V-14B-Turbo
* Wan-AI/Wan2.1-T2V-14B
If you are currently using any of the above models, it is recommended that you switch to other models as soon as possible to avoid any impact on your services.
### Platform Service Adjustment Notice
To further optimize resource allocation and provide more advanced and high-quality technical services, the platform will take the following models offline on July 3, 2025:
* Pro/deepseek-ai/DeepSeek-R1-0120
* Pro/deepseek-ai/DeepSeek-V3-1226
* Qwen/QwQ-32B-Preview
If you are currently using any of the above models, it is recommended that you switch to other models as soon as possible to avoid any impact on your services.
### Platform Maintenance Notice
To provide more rich, advanced, and high-quality services, the platform will undergo maintenance from 23:00 on June 10, 2025 to 08:00 on June 11, 2025.
Affected by the system maintenance:
1. The following functions on cloud.siliconflow\.cn will be suspended: registration, login, and other interface operations, including but not limited to the ones listed below.
* Online model experience, fine-tuning, and batch inference;
* Viewing model lists and details on the official website model marketplace;
* Online top-up, purchasing tiered packages, checking bills, and issuing invoices, etc.;
2. The `/user/info` API will be adjusted, and the fields `name`\ `image`\ `and` email will no longer be returned, with a fixed output of an empty string.
The platform's API services will not be affected by the maintenance and can be continuously used. We recommend that you check your account balance in advance to avoid service restrictions due to insufficient balance.
### Platform Service Adjustment Notice
SiliconFlow will initiate the `DeepSeek R1` model update.
For the `deepseek-ai/DeepSeek-R1` and `Pro/deepseek-ai/DeepSeek-R1` models, an "incremental" update to the latest `0528` version will be carried out.
After the update is completed, both models will be upgraded to the `0528` version. If needed, you can still use the older version of the model via `Pro/deepseek-ai/DeepSeek-R1-0120` until `June 28, 2025`, to achieve a smoother transition for your business.
### Platform Service Adjustment Notice
To further optimize resource allocation and provide more advanced and high-quality technical services, the platform will take the following models offline on June 5, 2025:
* Qwen/Qwen2-1.5B-Instruct
* Pro/Qwen/Qwen2-1.5B-Instruct
* Pro/Qwen/Qwen2-VL-7B-Instruct
* THUDM/chatglm3-6b
* internlm/internlm2\_5-20b-chat
* deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
* Pro/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
If you are currently using any of the above models, it is recommended that you switch to another model as soon as possible to avoid any impact on your service.
### Platform Service Adjustment Notice
To further optimize resource allocation and deliver more advanced and high-quality technical services, the platform will retire the HunyuanVideo model (non-HunyuanVideo-HD) on April 29, 2025.
If you are currently using this model, we strongly recommend switching to alternative models as soon as possible to avoid service disruptions.
### Platform Service Adjustment Notice
As of now, the `Pro/deepseek-ai/DeepSeek-V3` and `deepseek-ai/DeepSeek-V3` model has been updated to the latest 0324 version. You can still use the older model via `Pro/deepseek-ai/DeepSeek-V3-1226` to facilitate a smoother transition of your business.
### Platform Service Adjustment Notice
SiliconFlow will initiate the update of the DeepSeek V3 model.
For the `deepseek-ai/DeepSeek-V3` and `Pro/deepseek-ai/DeepSeek-V3` models, they will be "progressively" updated to the latest 0324 version.
After the update, both models will be at the 0324 version. If needed, you can still use the old version of the model via `deepseek-ai/DeepSeek-V3-1226` until April 30, 2025, to facilitate a smoother business transition.
### Platform Service Adjustment Notice
To better serve global developer users, SiliconFlow will soon launch an international site and gradually open multiple service regions.
Due to this adjustment, the existing api.siliconflow\.com API endpoint will be phased out at an appropriate time. Please switch to api.siliconflow\.cn as soon as possible to continue using the service.
We have already configured Global Traffic Manager (GTM) for the .cn endpoint to provide the same global access experience as the current .com endpoint. You only need to change the base URL of your API requests to api.siliconflow\.cn.
We recommend that you complete the migration by the end of this month (March 31). If you have any questions, please contact us at any time.
### Platform Service Adjustment Notice
To continuously improve user experience, the Rate Limits policy is being adjusted as follows:
Remove the RPH and RPD rate limits for deepseek-ai/DeepSeek-R1 and deepseek-ai/DeepSeek-V3.
As traffic and load change, the policy may be adjusted at any time, and Silicic Flow reserves the right to interpret.
### Platform Service Adjustment Notice
#### 1. Model Offline notice
To further optimize resource allocation and provide more advanced, high-quality, and compliant technical services, the platform will shut down certain models on March 6, 2025. The specific list of models involved is as follows:
* Chat models:
* AIDC-AI/Marco-o1
* meta-llama/Meta-Llama-3.1-8B-Instruct
* Pro/meta-llama/Meta-Llama-3.1-8B-Instruct
* meta-llama/Meta-Llama-3.1-70B-Instruct
* meta-llama/Meta-Llama-3.1-405B-Instruct
* meta-llama/Llama-3.3-70B-Instruct
* Image generation models:
* black-forest-labs/FLUX.1-schnell
* Pro/black-forest-labs/FLUX.1-schnell
* black-forest-labs/FLUX.1-dev
* black-forest-labs/FLUX.1-pro
* stabilityai/stable-diffusion-xl-base-1.0
* stabilityai/stable-diffusion-3-5-large
* stabilityai/stable-diffusion-3-5-large-turbo
* stabilityai/stable-diffusion-2-1
* deepseek-ai/Janus-Pro-7B
* Voice models:
* fishaudio/fish-speech-1.5
* FunAudioLLM/SenseVoiceSmall
* FunAudioLLM/CosyVoice2-0.5B
* fishaudio/fish-speech-1.4
* RVC-Boss/GPT-SoVITS
* Video models:
* Lightricks/LTX-Video
* genmo/mochi-1-preview
### Platform Service Adjustment Notice
To ensure the quality of platform services and the rational allocation of resources, the following adjustments to Rate Limits policies are now in effect:
1. Adjustments
New RPH Limit (Requests Per Hour, Per Hour Requests)
* Model Scope:deepseek-ai/DeepSeek-R1, deepseek-ai/DeepSeek-V3
* Applicable Users: All users
* Limit Standard: 30 requests/hour
2.New RPD Limit (Requests Per Day, Per Day Requests)
* Model Scope: deepseek-ai/DeepSeek-R1, deepseek-ai/DeepSeek-V3
* Applicable Users: Users who have not completed real-name authentication
* Limit Standard: 100 requests/day
Please note that these policies may be adjusted at any time based on traffic and load changes. Silicon Flowing Reserves the right to interpret these policies.
#### 1. Model Offline notice
To provide more stable, high-quality, and sustainable services, the following models will be offline on **February 27, 2025**:
* [01-ai/Yi-1.5-34B-Chat-16K](https://cloud.siliconflow.cn/models?target=01-ai/Yi-1.5-34B-Chat-16K)
* [01-ai/Yi-1.5-6B-Chat](https://cloud.siliconflow.cn/models?target=01-ai/Yi-1.5-6B-Chat)
* [01-ai/Yi-1.5-9B-Chat-16K](https://cloud.siliconflow.cn/models?target=01-ai/Yi-1.5-9B-Chat-16K)
* [stabilityai/stable-diffusion-3-medium](https://cloud.siliconflow.cn/models?target=stabilityai/stable-diffusion-3-medium)
* google/gemma-2-27b-it
* google/gemma-2-9b-it
* Pro/google/gemma-2-9b-it
If you are using any of these models, it is recommended to migrate to other models available on the platform as soon as possible.
### Platform service adjustment notice
#### DeepSeek-V3 model prices have been restored to the original price starting from Beijing time February 9, 2025, at 00:00.
Specific prices:
* Input: ¥2/M Tokens
* Output: ¥8/M Tokens
### Inference model output adjustment notice
The display of the reasoning chain in the inference model will be separated into a separate reasoning\_content field from the content field. This change is compatible with the OpenAI and DeepSeek API specifications, making it easier for various frameworks and upper-layer applications to trim the conversation in multi-round dialogues. For more details, see the Inference Model [(DeepSeek-R1) Usage](English/capabilities/reasoning).
### Platform service adjustment notice
#### Support for deepseek-ai/DeepSeek-R1 and deepseek-ai/DeepSeek-V3 Models
The specific pricing is as follows:
* `deepseek-ai/DeepSeek-R1` Input:￥4/ M Tokens Output: ￥16/ M Tokens
* `deepseek-ai/DeepSeek-V3`
* From February 1, 2025, to February 8, 2025, 24:00 Beijing Time, enjoy a limited-time discount price：Input：¥2￥1/ M Tokens Output：¥8￥2/ M Tokens，The original price will be restored from February 9, 2025, 00:00.
### Platform service adjustment notice
#### Image and Video URL Validity Period Adjusted to 1 Hour
To continue providing you with more advanced and high-quality services, the validity period of image and video URLs generated by large models will be adjusted to 1 hour starting from January 20, 2025.
If you are currently using the image and video generation service, please make sure to back up the files in time to avoid any business disruptions due to URL expiration.
### Platform service adjustment notice
#### LTX-Video Model Will Start Charging
To continue providing you with more advanced and high-quality services, the platform will start charging for video generation requests using the Lightricks/LTX-Video model starting from January 6, 2025, at a rate of 0.14 yuan per video.
### Platform service adjustment notice
#### 1. Model Offline notice
To provide more stable, high-quality, and sustainable services, the following models will be offline on December 19, 2024:
* [deepseek-ai/DeepSeek-V2-Chat](https://cloud.siliconflow.cn/models?target=deepseek-ai/DeepSeek-V2-Chat)
* [Qwen/Qwen2-72B-Instruct](https://cloud.siliconflow.cn/models?target=Qwen/Qwen2-72B-Instruct)
* [Vendor-A/Qwen/Qwen2-72B-Instruct](https://cloud.siliconflow.cn/models?target=Vendor-A/Qwen/Qwen2-72B-Instruct)
* [OpenGVLab/InternVL2-Llama3-76B](https://cloud.siliconflow.cn/models?target=OpenGVLab/InternVL2-Llama3-76B)
If you are using any of these models, it is recommended to migrate to other models available on the platform as soon as possible.
### Platform service adjustment notice
#### 1. Model offline notice
To provide more stable, high-quality, and sustainable services, the following models will be offline on December 13, 2024:
* [Qwen/Qwen2.5-Math-72B-Instruct](https://cloud.siliconflow.cn/models?target=Qwen/Qwen2.5-Math-72B-Instruct)
* [Tencent/Hunyuan-A52B-Instruct](https://cloud.siliconflow.cn/models?target=Tencent/Hunyuan-A52B-Instruct)
If you are using any of these models, it is recommended to migrate to other models available on the platform as soon as possible.
### Platform service adjustment notice
#### 1. Model offline notice
To provide more stable, high-quality, and sustainable services, the following models will be offline on November 22, 2024:
* [deepseek-ai/DeepSeek-Coder-V2-Instruct](https://cloud.siliconflow.cn/models?target=deepseek-ai/DeepSeek-Coder-V2-Instruct)
* [Qwen/Qwen2-57B-A14B-Instruct](https://cloud.siliconflow.cn/models?target=Qwen/Qwen2-57B-A14B-Instruct)
* [Pro/internlm/internlm2\_5-7b-chat](https://cloud.siliconflow.cn/models?target=Pro/internlm/internlm2_5-7b-chat)
* [Pro/THUDM/chatglm3-6b](https://cloud.siliconflow.cn/models?target=Pro/THUDM/chatglm3-6b)
* [Pro/01-ai/Yi-1.5-9B-Chat-16K](https://cloud.siliconflow.cn/models?target=Pro/01-ai/Yi-1.5-9B-Chat-16K)
* [Pro/01-ai/Yi-1.5-6B-Chat](https://cloud.siliconflow.cn/models?target=Pro/01-ai/Yi-1.5-6B-Chat)
If you are using any of these models, it is recommended to migrate to other models available on the platform as soon as possible.
#### 2.Email login method update
To further enhance service experience, the platform will update the login method starting from November 22, 2024: from the original "email account + password" method to an "email account + verification code" method.
#### 3. New Overseas API Endpoint
A new endpoint for overseas users has been added: [https://api-st.siliconflow.cn](https://api-st.siliconflow.cn). If you encounter network connection issues while using the original endpoint [https://api.siliconflow.cn](https://api.siliconflow.cn), it is recommended to switch to the new endpoint.
### Partial Model Pricing Adjustment Notice
To provide more stable, high-quality, and sustainable services, the [Vendor-A/Qwen/Qwen2-72B-Instruct model](https://cloud.siliconflow.cn/models?target=17885302571), which was previously offered for free, will start charging from October 17, 2024. The pricing details are as follows:
* Limited-time discount price：¥ 1.00 / M tokens
* Original price：¥ 4.13 / M tokens（the original price will be restored at a later date）