NVIDIA models

build.nvidia.com NIM catalog — OpenAI-compatible.

catalog updated 46d ago
build.nvidia.com hosts NIM microservices behind an OpenAI-compatible endpoint integrate.api.nvidia.com/v1. Generous free credits to start.
26 / 26
ModelMakerTypeContext$ In$ OutSpeed t/sCapabilityBest for
DeepSeek V4 Pro
deepseek-ai/deepseek-v4-pro-0813
DeepSeek AIchat1M$0.66$1.984596Frontier MoE (1.6T total / ~49B active, hybrid compressed-sparse attention) with ~1M context and three reasoning modes (Non-think / Think High / Think Max).
Kimi K3
moonshotai/kimi-k3
Moonshot AIchat1Mfreefree3895Moonshot's 2.78T-parameter KDA+MLA hybrid MoE, natively multimodal at 1M context — the largest model in this catalog. Listed production on NIM 2026-08-20 with a free endpoint and downloadable weights.
NVIDIA Nemotron 3 Ultra 550B-A55B
nvidia/nemotron-3-ultra-550b-a55b
NVIDIAchat1Mfreefree7092Frontier-tier open MoE reasoning (550B total / 55B active, hybrid Mamba-2 + LatentMoE) for maximum accuracy on hard reasoning, math and agentic decision-making.
Llama 3.1 Nemotron Ultra 253B v1
nvidia/llama-3.1-nemotron-ultra-253b-v1
NVIDIAchat128k——4592Prior-gen high-accuracy open reasoning, agentic tool-calling, RAG and complex math/coding; derivative of Llama-3.1-405B compressed via Neural Architecture Searc
Qwen3.5 397B-A17B
qwen/qwen3.5-397b-a17b
Alibaba Cloud (Qwen)chat262kfreefree4591Qwen's 397B/17B-active MoE flagship on NIM — 262k native context that NVIDIA documents as extensible to 1,010,000 via YaRN RoPE scaling. The largest Qwen text model available here.
GLM-5.2
z-ai/glm-5.2
Z.ai (Zhipu)chat1Mfreefree4591The base model GLM-5.3 was post-trained from, at a documented 1,000,000 tokens both in and out — the largest output ceiling in this catalog. Replaces the deprecated glm-5.1 NIM.
Mistral Large 3 675B
mistralai/mistral-large-3-675b-instruct-2512
Mistral AIchat262kfreefree4589Mistral's 675B MoE vision-language flagship. The strongest European-sovereign option in this catalog and the best multilingual scores of the frontier tier.
Kimi K2.6
moonshotai/kimi-k2.6
Moonshot AIchat262kfreefree5088Replaces the retired kimi-k2-instruct on NIM. General-purpose agentic model a tier below K3, and the one Cloudflare also serves per-token — useful for comparing a credit-based and a priced path to the same weights.
NVIDIA Nemotron 3 Super 120B-A12B
nvidia/nemotron-3-super-120b-a12b
NVIDIAchat262kfreefree15087Newest-gen hybrid Mamba-2 + LatentMoE reasoning (120B total / 12B active, first Nemotron pre-trained in NVFP4) with up to ~1M context for deep document reasonin
Qwen3.5 122B-A10B
qwen/qwen3.5-122b-a10b
Alibaba Cloud (Qwen)chat262kfreefree6285The mid-size Qwen3.5 MoE — 122B total, 10B active. A cheaper stop between Qwen3.5 397B and the small dense models.
Muse Glimmer 30B
meta/muse-glimmer-30b
Meta Superintelligence Labschat131kfreefree8583The open, Apache-2.0 half of Meta's post-Llama Muse line: a 29.8B dense multimodal reasoner distilled from Muse Spark, with native tool calling and a separate reasoning output. Deliberately dense at a moment when its peers are all sparse.
Mistral Medium 3.5 128B
mistralai/mistral-medium-3.5-128b
Mistral AIchat262kfreefree6282Dense 128B — no MoE routing, so latency is more predictable under load than a sparse model of similar quality.
Nemotron 3.5 Lightning 30B-A3B preview
nvidia/nemotron-3.5-lightning-30b-a3b
NVIDIAchat1Mfreefree20080NVIDIA's new 3.5 generation: 30B total / 3B active interleaving Mamba-2, attention and MoE blocks across 52 layers, with speculative decoding, built for long-running sub-agent roles. Preview on NIM — gate it accordingly.
Laguna XS 2.1
poolside/laguna-xs-2.1
Poolsidechat262kfreefree1208033B MoE / 3B active aimed squarely at agentic coding and terminal work — a specialist rather than a general model.
GLM-4.7
z-ai/glm-4.7
Z.ai (Zhipu)chat131kfreefree6480The previous GLM generation, still hosted — a cheaper fallback when GLM-5.2's context is not needed.
Mistral Small 4 119B
mistralai/mistral-small-4-119b-2603
Mistral AIchat262kfreefree7278Hybrid MoE and multimodal despite the 'Small' name — 119B total at 262k context.
Gemma 4 31B
google/gemma-4-31b-it
Googlechat256kfreefree8077The dense 31B Gemma 4, the model Meta benchmarks Muse Glimmer against.
gpt-oss-20b
openai/gpt-oss-20b
OpenAIchat131kfreefree11076Smaller OpenAI open-weight MoE for cost-efficient reasoning and agentic tasks; good for latency-sensitive or local-friendly deployments.
NVIDIA Nemotron 3 Nano 30B-A3B
nvidia/nemotron-nano-3-30b-a3b
NVIDIAchat256kfreefree24074Efficient open hybrid Mamba-2 + MoE reasoning (30B total / 3B active) with up to ~1M context and configurable thinking budget; ~4x throughput of Nemotron 2 Nano
NVIDIA Nemotron 3 Nano Omni 30B-A3B (Reasoning)
nvidia/nemotron-3-nano-omni-30b-a3b-reasoning
NVIDIAchat256kfreefree24073Multimodal perception sub-agent for agentic AI: native text, image, video and audio input with reasoning.
DiffusionGemma 26B-A4B
google/diffusiongemma-26b-a4b-it
Googlechat262kfreefree1.1k73A diffusion language model that decodes 256-token blocks in parallel rather than token by token — NVIDIA reports over 1,100 tok/s on an H100. A genuinely different latency profile from everything else in this catalog, and the only diffusion LM in it.
Ministral 14B
mistralai/ministral-14b-instruct-2512
Mistral AIchat262kfreefree8668The small end of the Mistral 3 generation — a 14B vision-language model at full 262k context, for high-volume multilingual work.
DeepSeek V4 Flash
deepseek-ai/deepseek-v4-flash-0731
DeepSeek AIchat1M$0.08$0.168066Fast, cost-efficient MoE with ~1M context optimized for high-throughput coding and agentic workflows; the latency-oriented sibling of V4 Pro.
Nemotron 3 Embed 1B
nvidia/nemotron-3-embed-1b
NVIDIAembedding33kfreefree—262,048-dimension embedding NIM covering 34 languages — a newer alternative to the llama-3.2 embedqa NIMs already listed.
Llama Nemotron Embed VL 1B v2
nvidia/llama-nemotron-embed-vl-1b-v2
NVIDIAembedding8kfreefree—26Multimodal (text + image) embedding NIM, and the live replacement for the retired llama-3.2-nv-embedqa-1b-v2. Pairs with nemotron-3-embed-1b for text-only work.
Llama Nemotron Rerank VL 1B v2
nvidia/llama-nemotron-rerank-vl-1b-v2
NVIDIArerank8k———25Vision-language reranker for multimodal/document RAG; reorders text-and-image candidates by relevance for document-extraction and visual retrieval pipelines. Retained deliberately: /v1/models does not enumerate rerank endpoints, so its absence there is not evidence of removal — its catalog page is live with no deprecation notice (checked 2026-09-05).
Sign in to continue

LLM Switchboard is private — sign in with Authlee to access the control room.

Sign in with Authlee
← Back to home