硅基流动（SiliconFlow）提供基于优秀开源模型的生成式AI云服务。主要能力包括文本对话、图像生成、视频生成和语音合成。该平台强调这些AI生成任务的高性价比。典型用例涉及利用这些多模态AI服务进行各种创意和对话应用。
基于开源模型提供高性价比的生成式AI云服务，支持文本对话、图像生成、视频生成和语音合成。
cloud.siliconflow.cnapi.siliconflow.cncloud.siliconflow.comapi.siliconflow.com完成站点认证，解锁以下权益：
排名基于社区提交的测试数据与定期健康探测，仅供参考，非官方数据。
硅基流动（SiliconFlow）提供基于优秀开源模型的生成式AI云服务。主要能力包括文本对话、图像生成、视频生成和语音合成。该平台强调这些AI生成任务的高性价比。典型用例涉及利用这些多模态AI服务进行各种创意和对话应用。
按 62 个模型行对比最新 audit、最新测速、吞吐、延迟与按 token 计费价格。
| 模型 | 输入 ($/M) | 输出 ($/M) | 检测 | 速度 | 延迟 |
|---|---|---|---|---|---|
| - | - | — | 52.1 t/s | 10.99 s | |
| - | - | 888486100 | 83.1 t/s | 1.16 s | |
Qwen/Qwen3.6-27B | - | - | — | 36.1 t/s | 41.10 s |
| - | - | 846880100 | 38.4 t/s | 0.49 s | |
| - | - | — | 73.4 t/s | 15.12 s | |
Qwen/Qwen3.6-35B-A3B | - | - | — | 115.8 t/s | 8.56 s |
| - | - | — | 49.6 t/s | 9.04 s | |
ByteDance-Seed/Seed-OSS-36B-Instruct | - | - | — | 61.6 t/s | 12.78 s |
| - | - | — | 89.9 t/s | 6.43 s | |
PaddlePaddle/PaddleOCR-VL-1.5 | - | - | — | 279.6 t/s | 4.15 s |
| - | - | — | 34.2 t/s | 33.63 s | |
| - | - | — | 136.6 t/s | 6.06 s | |
| - | - | — | 32.8 t/s | 29.36 s | |
| - | - | — | 95.6 t/s | 16.29 s | |
Qwen/Qwen3-Omni-30B-A3B-Instruct | - | - | — | 128.8 t/s | 0.47 s |
| - | - | — | 54.7 t/s | 7.54 s | |
| - | - | — | 71.4 t/s | 9.63 s | |
| - | - | — | 43.6 t/s | 27.77 s | |
| - | - | — | 84.2 t/s | 3.37 s | |
| - | - | — | 44.6 t/s | 14.30 s | |
| - | - | — | 62.3 t/s | 16.76 s | |
| - | - | — | 74.7 t/s | 16.10 s | |
deepseek-ai/DeepSeek-R1-0528-Qwen3-8B | - | - | — | 48.5 t/s | 27.88 s |
deepseek-ai/DeepSeek-R1-Distill-Llama-70B | - | - | — | 22.6 t/s | 6.14 s |
| - | - | — | 15.5 t/s | 0.96 s | |
| - | - | — | 14.9 t/s | 2.66 s | |
inclusionAI/Ling-1T | - | - | — | 12.5 t/s | 1.10 s |
internlm/internlm2_5-7b-chat | - | - | — | 69.3 t/s | 0.57 s |
LoRA/Qwen/Qwen2.5-14B-Instruct | - | - | — | 58.1 t/s | 1.02 s |
LoRA/Qwen/Qwen2.5-32B-Instruct | - | - | — | 64.7 t/s | 0.84 s |
LoRA/Qwen/Qwen2.5-72B-Instruct | - | - | — | 24.6 t/s | 3.37 s |
Pro/Qwen/Qwen2-7B-Instruct | - | - | — | 83.8 t/s | 0.60 s |
Pro/Qwen/Qwen2.5-Coder-7B-Instruct | - | - | — | 28.3 t/s | 0.63 s |
| - | - | — | 73.5 t/s | 0.71 s | |
Pro/THUDM/GLM-4.1V-9B-Thinking | - | - | — | 65.0 t/s | 15.30 s |
Qwen/Qwen2.5-72B-Instruct | - | - | — | 32.5 t/s | 0.79 s |
Qwen/Qwen2.5-72B-Instruct-128K | - | - | — | 20.3 t/s | 0.90 s |
Qwen/Qwen2.5-7B-Instruct | - | - | — | 39.4 t/s | 1.40 s |
Qwen/Qwen2.5-Coder-32B-Instruct | - | - | — | 23.6 t/s | 1.18 s |
Qwen/Qwen2.5-VL-72B-Instruct | - | - | — | 28.1 t/s | 0.92 s |
Qwen/Qwen3-235B-A22B-Instruct-2507 | - | - | — | 21.0 t/s | 1.49 s |
Qwen/Qwen3-Coder-30B-A3B-Instruct | - | - | — | 38.2 t/s | 0.78 s |
Qwen/Qwen3-VL-8B-Instruct | - | - | — | 106.5 t/s | 0.92 s |
tencent/Hunyuan-MT-7B | - | - | — | 46.6 t/s | 1.32 s |
| - | - | — | 72.0 t/s | 0.95 s | |
| - | - | — | 74.3 t/s | 0.60 s | |
THUDM/GLM-4.1V-9B-Thinking | - | - | — | 41.2 t/s | 23.48 s |
THUDM/GLM-Z1-9B-0414 | - | - | — | 176.0 t/s | 13.40 s |
Vendor-A/Qwen/Qwen2.5-72B-Instruct | - | - | — | 27.7 t/s | 1.25 s |
| - | - | 10010086100 | 21.4 t/s | 5.27 s | |
| - | - | — | 20.9 t/s | 8.59 s | |
| - | - | — | 34.4 t/s | 22.51 s | |
| - | - | — | 24.2 t/s | 0.98 s | |
| - | - | — | 60.6 t/s | 15.38 s | |
| - | - | — | 73.0 t/s | 6.15 s | |
| - | - | — | 85.7 t/s | 0.39 s | |
| - | - | — | 29.0 t/s | 29.09 s | |
| - | - | — | 22.5 t/s | 9.04 s | |
| - | - | — | 15.0 t/s | 62.39 s | |
| - | - | — | 67.7 t/s | 10.66 s | |
| - | - | — | 11.5 t/s | 60.01 s | |
| - | - | — | 68.6 t/s | 9.89 s |
当前显示 62 / 62 个模型行
| 时间 | 模型 | 速度 | 延迟 |
|---|---|---|---|
| Jun 21, 08:27 PM | zai-org/GLM-5.2 | 52.06 tok/s | 10.99s |
| Jun 18, 05:23 AM | Qwen/Qwen3.6-27B | 36.08 tok/s | 41.10s |
| Jun 15, 02:02 AM | deepseek-ai/DeepSeek-V3.2 | 17.94 tok/s | 2.29s |
| Jun 14, 04:02 PM | Qwen/Qwen3.5-35B-A3B | 136.58 tok/s | 6.06s |
| Jun 14, 04:01 PM | nex-agi/Nex-N2-Pro | 83.12 tok/s | 1.16s |
| May 28, 01:20 PM | deepseek-ai/DeepSeek-V4-Flash | 68.08 tok/s | 2.00s |
| May 28, 01:19 PM | deepseek-ai/DeepSeek-V3.2 | 23.22 tok/s | 1.05s |
| May 24, 10:53 AM | deepseek-ai/deepseek-v4-flash | 38.38 tok/s | 0.49s |
| May 20, 03:41 AM | deepseek-ai/DeepSeek-V4-Flash | 37.88 tok/s | 43.15s |
| May 11, 12:11 PM | Qwen/Qwen3.5-4B | 99.84 tok/s | 9.04s |
用 358 个 LMSpeed 信号，把 SiliconFlow 的 6 个相近 API 替代服务商放在一起比较：共享模型覆盖、价格、实测速度、可用性和免费模型。
| 服务商 | 对比理由 | 模型数 | 免费项 | 均价 | 速度 | 30 天可用性 |
|---|---|---|---|---|---|---|
| SiliconFlow siliconflow Provides cost-effective generative AI cloud services based on open-source models for text, image, video, and audio generation. | 当前服务商基线 | 60 | 0 | N/A | 53 tok/s | 79.9% |
| OpenRouter openrouter A unified API interface providing access to over 300 models from 60+ providers, including OpenAI, Anthropic, and Google. |
| 221 | 0 | N/A | 86 tok/s | 99.7% |
nvidia-nim NVIDIA NIM provides optimized AI model inference APIs for LLMs, vision, and embedding models through NVIDIA cloud infrastructure. |
| 53 | 0 | N/A | 66 tok/s | 99.4% |
api-kriora-com Provides OpenAI-compatible APIs and managed GPU instances for deploying and scaling open-source AI models. |
| 9 | 0 | N/A | 565 tok/s | 99.6% |
api-fireworks-ai Fireworks AI provides a cloud platform for running and fine-tuning open-source AI models with optimized inference for production applications. |
| 5 | 0 | N/A | 139 tok/s | 99.8% |
550c-cloud 共绩算力 (550c.cloud) is a shared computing platform providing Ollama-hosted open-source AI model inference via OpenAI-compatible API. |
| 4 | 0 | N/A | 112 tok/s | 82.6% |
www-sophnet-com An AI development platform offering API access to models like DeepSeek and Qwen for tasks such as text generation and code creation. |
| 3 | 0 | N/A | 133 tok/s | 99.6% |
