DeepSeek (@deepseek_ai) V4.1 Flash 1017 is now available on NEAR (@NEARProtocol) $NEAR AI Cloud, expanding the model catalog of the AI inference platform. Near AI (@near_ai) says the model is
DeepSeek (@deepseek_ai) V4.1 Flash 1017 is now available on NEAR (@NEARProtocol) $NEAR AI Cloud, expanding the model catalog of the AI inference platform. Near AI (@near_ai) says the model is better at coding, reasoning and agentic tasks, with faster completion and shorter outputs.
Pricing starts at $0.30 per million input tokens and $1.20 per million output tokens. The model runs on partner infrastructure rather than inside a private trusted execution environment (TEE). That distinction matters for users who choose NEAR AI Cloud mainly for privacy.
According to NEAR AI's model listing, as referenced in a public models(.)dev repository change, the route is served through OpenRouter using zero data retention inference endpoints. The same listing shows a cache-read price of $0.006 per million tokens.
NEAR AI Cloud markets itself as a private inference platform. Its website says TEE-hosted open-weight models return a hardware-signed proof with every response. The DeepSeek V4.1 Flash 1017 route is a different tier from those private options, so users should check the privacy level of each model before sending sensitive data.
DeepSeek's change log says DeepSeek V4.1 Flash has been upgraded to DeepSeek-V4.1-Flash-1017 on October 17, 2026. The company adds that the API calling method and pricing remain unchanged on its own platform.
On architecture, DeepSeek's Hugging Face model card describes V4.1 Flash as a multimodal Mixture-of-Experts model with 552B backbone parameters and support for contexts of up to one million tokens. It says the design activates only 8B parameters per token during prefill and 16B during decode, which is aimed at cutting costs for input-heavy agentic workloads.
NEAR AI Cloud launched in December 2025, according to Coinspeaker via Yahoo Finance, with every request processed by Intel TDX and NVIDIA Confidential Computing hardware. Adding third-party-served models such as this one widens developer choice, though with different privacy guarantees.
Developers can access the model through the platform's OpenAI-compatible API. Current availability, context limits and pricing are listed on the NEAR AI model catalog.
