SiliconFlow's official release notes document several platform changes affecting developers using the api.siliconflow.cn/v1 endpoint. On 2026-08-25, the provider announced that DeepSeek-V4-Flash will move to a time-of-day pricing model effective 2026-09-01, charging ¥0.15/M (cache hit), ¥1.5/M (input), and ¥4.5/M (outp Earlier 2026 pricing changes are also itemized: a new ¥1/M cache-hit input rate for DeepSeek-V4-Pro from 2026-08-03, and 2026-06-30 adjustments covering Nex-N2-Pro moving from a free trial to paid (¥1.75/M input, ¥7/M output, ¥0.175/M cache hit), DeepSeek-V4-Pro discount expiration back to ¥12/M input, ¥24/M output, ¥0

SiliconFlow (China) is listed on Phaseo as a paid API provider for Qwen 3 VL 8B Instruct, alongside AtlasCloud, NovitaAI, and Together. The excerpt shows SiliconFlow listing prices of $0.18 per 1M input tokens and $0.68 per 1M output tokens for the multimodal vision-language model, which is higher than AtlasCloud and N This entry confirms SiliconFlow (China) as an active hosting route for Qwen3-VL series models via its API at api.siliconflow.cn, with current list pricing recorded in recent Phaseo snapshots for multimodal inference. However, the page notes no gateway telemetry is yet available, and benchmark scores reflect the underly