NVIDIA NIM
NVIDIA NIM（微服务推理平台）提供经 NVIDIA Blackwell / Hopper GPU 深度优化的企业级与前沿开源大模型 API。注册 NVIDIA 开发者账号即赠 1,000 免费推理积分，支持 OpenAI 兼容协议直连。
BEVFormer Bird's-Eye-View Perception
NVIDIA BEVFormer bird's-eye-view camera fusion perception model on NIM.
SparseDrive End-to-End Perception
NVIDIA SparseDrive end-to-end perception and motion forecasting on NIM.
StreamPETR 3D Perception
NVIDIA StreamPETR real-time 3D object detection microservice on NIM.
Synthetic Video Detector (SVD)
NVIDIA AI synthetic deepfake video detection microservice on NIM.
Cosmos Transfer 1.0 7B
NVIDIA Cosmos Transfer 1.0 7B high-fidelity physical video transfer on NIM.
Cosmos Transfer 2.5 2B
NVIDIA Cosmos Transfer 2.5 2B physical video style transfer on NIM.
Cosmos 3 Nano Reasoner
NVIDIA Cosmos 3 Nano physical spatial dynamics reasoning model on NIM.
Cosmos 3 Nano World Model
NVIDIA Cosmos 3 Nano physical AI world foundation model on NIM.
PaliGemma 3B Vision-Language
Google's 3B vision-language model for image captioning and OCR on NIM.
DiffusionGemma 26B-A4B IT
Google's Diffusion-Transformer hybrid image synthesis model on NIM.
Nemotron 3 Embed 1B
NVIDIA's 1B dense text embedding model with top MTEB retrieval performance.
Riva Translate 4B Instruct v1.1
NVIDIA Riva 4B high-throughput machine translation v1.1 model on NIM.
Riva Translate 4B Instruct v2
NVIDIA Riva 4B neural machine translation v2 model on NIM.
NVIDIA Active Speaker Detection
NVIDIA audio-visual neural active speaker detection microservice.
NVIDIA Background Noise Removal
NVIDIA real-time background noise removal and echo cancellation.
NVIDIA Studio Voice
NVIDIA studio-quality voice enhancement and audio restoration microservice.
Nemotron VoiceChat Duplex
NVIDIA's full-duplex sub-300ms real-time conversational voice agent on NIM.
Magpie TTS Zero-Shot
NVIDIA's neural zero-shot voice cloning and TTS model on NIM.
Llama Guard 4 12B
Meta's 4th gen multimodal safety & content moderation model on NIM.
Llama 3.1 Nemotron Safety Guard 8B v3
NVIDIA 8B safety guard model for prompt injection & jailbreak detection.
Nemotron 3.5 Content Safety
NVIDIA NeMo Guardrails content safety & moderation model on NIM.
Ising Calibration 1.0 35B-A3B
NVIDIA 35B-A3B MoE reasoning calibration model on NIM.
Ising Calibration 1.5 31B
NVIDIA 31B reasoning calibration and verification model on NIM.
Laguna XS 2.1
Poolside's specialized coding intelligence model on NVIDIA NIM.
Llama 3.2 90B Vision Instruct
Meta's high-capacity 90B multimodal vision model on NVIDIA NIM.
Llama 3.2 11B Vision Instruct
Meta's lightweight 11B multimodal vision model on NVIDIA NIM.
Gemma 4 31B IT
Google's 2026 Gemma 4 31B instruction-tuned model on NVIDIA NIM.
GPT-OSS 20B
OpenAI's fast 20B open-weights model on NVIDIA NIM for code and agent loops.
GPT-OSS 120B
OpenAI's 120B open-weights model hosted on NVIDIA NIM with 1M context.
Mistral Nemotron
Mistral AI and NVIDIA jointly optimized model for structured JSON and reasoning.
Nemotron 3 Nano Omni
NVIDIA 30B-A3B omni multimodal vision-language reasoning model.
Nemotron 3 Nano 30B
NVIDIA ultra-efficient 30B MoE (3B active) low-latency coding model.
Nemotron 3 Super 120B
NVIDIA 120B MoE (12B active) high-throughput reasoning model on NIM.
Nemotron 3 Ultra Free
NVIDIA flagship 550B MoE (55B active) reasoning & code synthesis model via NIM.
MiniMax M3
MiniMax M3 100 万 Token MSA 稀疏注意力旗舰推理模型，通过 NVIDIA NIM 托管，SWE-bench Pro 59% 顶级编程表现。
Meta Muse Glimmer 30B
Meta 2026 原生 30B 图文多模态推理大模型，NVIDIA NIM 官方托管，支持图片输入、工具调用与独立思维链。
Nemotron 3.5 Lightning
NVIDIA 官方 30B-A3B Mamba-2/MoE 混合架构（2026 年 8 月发布），NVFP4 硬件量化，1M 上下文，毫秒级响应。
DeepSeek V4 Flash Free
DeepSeek V4 Flash 284B 极速 MoE（0731 最新版）通过 NVIDIA NIM 托管，1M 上下文，专为代码补全与 Agent 工具调用优化。
DeepSeek V4 Pro
DeepSeek V4 Pro 1.65T MoE 旗舰（0813 最新版）通过 NVIDIA NIM 托管，100 万 Token 上下文，注册即赠 1,000 免费积分。
Kimi K3
NVIDIA NIM 官方托管的 Moonshot AI Kimi K3（2.8T MoE 旗舰，1M 上下文，原生视觉多模态），40 RPM 免费限流，免绑卡可用。
在浏览器里试用
选个模型、粘贴你自己的免费 Key，即可运行。你的 Key 只用于调用一次该提供商，绝不存储在我们的服务器上。
选个模型、粘贴你自己的免费 Key，即可运行。你的 Key 只用于调用一次该提供商，绝不存储在我们的服务器上。
社区反馈
评论与你的 GitHub 账号绑定，登录后即可参与讨论。