One API for all models. Search our library, deploy and run inference on NVIDIA GPUs in seconds
Unlock $1 free API credit on first recharge - generate up to ~4M tokens
Kimi K2.5 is Moonshot AI's most powerful open-source model to date — a native multimodal agentic model built through continual pretraining on 15 trillion mixed visual and text tokens atop Kimi-K2-Base. With 1T total parameters (32B active), it seamlessly integrates vision, language, and advanced agentic capabilities including an Agent Swarm paradigm that coordinates up to 100 parallel sub-agents, reducing execution time by 4.5x on parallelizable tasks.
MiniMax-M2.1 is a SOTA open-source coding and agentic model with 230B total parameters and only 10B active per token (23:1 sparsity ratio). Released December 2025, it achieves 74% on SWE-bench Verified — competitive with Claude Sonnet 4.5 — at a fraction of the cost. It excels in multilingual code development (Python, Java, Go, Rust, C++, TypeScript, Kotlin), long-horizon agentic workflows, and office automation.
GLM-4.7 is Z.ai's (formerly Zhipu AI) new-generation flagship model with 355B total parameters and 32B activated per forward pass. It introduces Interleaved Thinking, Preserved Thinking, and Turn-level Thinking — enabling models to reason before actions and maintain coherent state across long coding sessions. Achieves 95.7% on AIME 2025, 73.8% on SWE-bench, and 87.4% on τ²-Bench.
Kimi K2 Thinking is the first open-weights model to achieve SOTA performance against leading closed-source models (GPT-5, Claude 4.5 Sonnet) across major benchmarks including HLE (44.9%), BrowseComp (60.2%), and SWE-Bench Verified (71.3%). Built on a 1T parameter MoE architecture with 32B active per token and native INT4 quantization via QAT, it maintains stable tool-use across 200–300 sequential calls within a 256K context window.
Qwen3.5-Flash is the production-hosted, closed-source API version of the Qwen3.5-35B-A3B model, served via Alibaba Cloud Model Studio. It delivers frontier-adjacent intelligence at roughly 1/13th the cost of Claude Sonnet 4.6, with a 1M token context window, native tool calling, built-in web search, and code interpreter support. Responses are 6x faster than Claude Sonnet 4.6 with competitive quality on agentic benchmarks.
Qwen3.5-27B is a dense (non-MoE) transformer model and the only full-weight model in the Qwen3.5 Medium Series. Released February 24, 2026, it achieves 72.4% on SWE-bench Verified — matching GPT-5 mini — despite having just 27B parameters. It supports native multimodal input (text + images + video) via early fusion, runs on a 22GB Mac M-series device, and natively extends to 1M token contexts.
Qwen3.5-35B-A3B is the breakout model of the Qwen3.5 Medium Series and arguably the biggest efficiency breakthrough in recent open-source AI. Despite having only 3B active parameters per token (8.6% of total), it outperforms the previous generation's 235B model on most benchmarks, as well as GPT-5 mini and Claude Sonnet 4.5 on knowledge (MMMLU) and visual reasoning (MMMU-Pro). It runs on an 8GB GPU and supports 256K context natively.
Qwen3.5-122B-A10B is the most powerful open-source model in the Qwen3.5 Medium Series. With 122B total parameters and 10B active per token across a 48-layer hybrid architecture, it delivers the strongest knowledge, vision, and function-calling performance in the medium class — scoring 86.6% on GPQA Diamond (beating GPT-5 mini's 82.8%), 72.2% on BFCL-V4 tool calling (vs GPT-5 mini's 55.5%), 92.1% on OCRBench, and 83.9% on MMMU. Supports text, image, and video input natively via early fusion.
Qwen3.5-397B-A17B is Alibaba's flagship open-source model — the first in the Qwen3.5 series (released February 16, 2026) and the most capable open-weight model in the family. It is a native multimodal model trained from scratch on trillions of text, image, and video tokens using early fusion across 201 languages. With 397B total parameters and 17B active per token, it outperforms all Qwen3-VL models on vision tasks while matching or exceeding text-only frontiers. The hosted version is called Qwen3.5-Plus.
DeepSeek-V3.2 is DeepSeek's frontier open-source model with 685B total parameters and novel DeepSeek Sparse Attention (DSA) that reduces long-context computational cost by 50%. Trained with a scalable RL framework, it achieves performance comparable to GPT-5, earning gold-medal results at the 2025 IMO and IOI. The model includes reasoning and tool-use capabilities through large-scale agentic synthesis.
DeepSeek-R1-0528 is the May 2025 update to the original DeepSeek-R1. Built on the DeepSeek-V3 backbone with 671B total parameters and 37B active per inference pass (MoE), it achieves performance on par with OpenAI o1. Key improvements include 87.5% on AIME 2025 (up from 70%), reduced hallucinations, enhanced front-end capabilities, and newly added JSON output and function calling support.
Qwen3-Max is Alibaba Cloud's most powerful closed-source model in the Qwen3 series, featuring a 235B Mixture-of-Experts architecture with 22B parameters active per forward pass. It delivers frontier-level performance in complex reasoning, multilingual tasks, long-context understanding, and advanced coding — rivaling GPT-4o and Claude Sonnet on major benchmarks. Accessible exclusively via the DashScope API.
Qwen3-VL-235B-A22B-Thinking is the most powerful vision-language model in the Qwen series. With 235B total parameters (22B active per token) and deep reasoning capabilities in thinking mode, it excels in multimodal STEM/math reasoning, visual agent tasks, GUI automation, spatial perception, long video comprehension, and multilingual OCR across 32 languages.
Qwen3-Coder-480B-A35B-Instruct is Alibaba's flagship open-source coding model powered by a sparse Mixture-of-Experts (MoE) architecture with 480B total parameters and 35B activated per forward pass. It achieves state-of-the-art (SOTA) performance among open-source models, supporting up to 256K context. Ideal for agentic coding, complex refactoring, and large-scale software engineering.
Qwen3-Next-80B-A3B-Thinking is a next-generation foundation model from Alibaba's Qwen team featuring a revolutionary Hybrid Attention mechanism (Gated DeltaNet + Gated Attention) with High-Sparsity MoE architecture. With 80B total parameters and only 3.9B active per token, it delivers 10x higher throughput than Qwen3-32B on long contexts while outperforming Gemini-2.5-Flash-Thinking on multiple benchmarks.
Nemotron 3 Nano 30B-A3B is NVIDIA's flagship open reasoning model, featuring a revolutionary hybrid Mamba-Transformer Mixture-of-Experts architecture. With 31.6B total parameters but only 3.2B active per forward pass, it delivers up to 3.3× higher throughput than comparable models while achieving state-of-the-art accuracy on reasoning, coding, and agentic benchmarks. The model supports up to 1M token context length and features configurable reasoning depth with thinking budget control.
Llama 3.3 70B Instruct is a 70B-parameter open-weight large language model from Meta, optimized for instruction following, complex reasoning, and multi-turn conversations.It is well suited for enterprise use cases such as advanced chat assistants, code reasoning, and long-document analysis with large context windows.
Released in late 2025, HunyuanOCR is an open-source contribution from Tencent that outperforms many larger proprietary models. It utilizes a "Global-to-Local" architecture with a SigLIP-v2 visual encoder to handle high-resolution inputs and extreme aspect ratios (like long receipts) without splitting images artificially.
DeepSeek R1 Distill LLaMA 70B is optimized for efficient, high-level reasoning and conversational intelligence. It delivers near frontier-level analytical performance while running on significantly smaller hardware.
Fara 7B is a compact and efficient transformer model developed by Microsoft for high-speed inference, instruction following, text generation, and lightweight reasoning tasks. Its small parameter size allows easy deployment on consumer GPUs and edge devices while maintaining strong performance.
NVIDIA Orchestrator is purpose-built for agent workflows and complex task sequencing. It excels in planning, structured reasoning, autonomous execution, and coordinating multiple tools or APIs. With deep GPU-level optimization, it delivers superior throughput and low latency in enterprise automation scenarios.
Qwen3-Coder-30B-A3B-Instruct is a sparse Mixture-of-Experts (MoE) model with around 30.5B total parameters (3.3B active per inference), 48 layers, supporting extremely long context (native 262,144 tokens — extendable to 1M in some deployments).
Introducing gpt-oss-120B, OpenAI's flagship open-weight model in the gpt-oss series, built for advanced reasoning, large-scale agentic workloads, and enterprise-grade automation. With 120B parameters and a highly optimized Mixture-of-Experts (MoE) architecture, it activates 12B parameters during inference, delivering exceptional intelligence while maintaining competitive latency. Designed for complex reasoning, multi-task agents, and long-horizon planning, gpt-oss-120B brings frontier-level capability to commercial and self-hosted deployments.
Qwen3-VL-30B-A3-Instruct is a large-scale, high-capacity vision-language instruction model designed for advanced multimodal reasoning. It delivers significantly stronger visual understanding, OCR accuracy, document reasoning, long-context comprehension, and agent-style interactions compared to smaller Qwen-VL variants.
Qwen3-VL is a vision-language instruction-tuned model capable of understanding text and images. It supports streaming, OCR, and rich multimodal conversations.
Whisper V3 delivers high-accuracy speech recognition across 99 languages. Ideal for transcription, subtitles, and accessibility.
The Qwen3 series VL models has been comprehensively upgraded in areas such as visual coding and spatial perception. Its visual perception and recognition capabilities have significantly improved, supporting the understanding of ultra-long videos, and its OCR functionality has undergone a major enhancement.
Faster, lighter vision model for real-time use cases.
Vision-language model that understands images and text together.
Lightweight, fast coding model optimized for speed.
Balanced general-purpose model for everyday chat and analysis tasks.
Powered by Qwen3, this is a powerful Coding Agent that excels in tool calling and environment interaction to achieve autonomous programming. It combines outstanding coding proficiency with versatile general-purpose abilities.
Welcome to the gpt-oss series, OpenAI's open-weight models designed for powerful reasoning, agentic tasks, and versatile developer use cases. gpt-oss-20b is a 21.5B parameter model with Mixture-of-Experts (MoE) architecture, featuring 3.6B active parameters during inference. It's optimized for lower latency and local or specialized use-cases, supporting configurable reasoning depth for agentic applications.
Mistral 7B is a 7.3B parameter language model celebrated for its efficiency, outperforming larger models on many benchmarks. The v0.3 instruct version is specifically fine-tuned for chat and instruction-following tasks.
P-Image is Pruna's ultra-fast text-to-image generation model with automatic prompt enhancement and 2-stage refinement. It delivers state-of-the-art AI images in less than one second per image with strong prompt adherence and high visual quality.
P-Image Edit enables professional-grade image editing and compositing with extremely fast inference times. It supports multi-image conditioning and precise instruction-following for production workflows.
This model generates and edits images from text prompts using a Latent Diffusion framework, with dual CLIP text encoders.
Z-Image Turbo is built for ultra-fast image generation, ideal for low-latency workflows and real-time creative tasks.
P-Video is a premium AI video generation model designed for cinematic-quality outputs. It supports text-to-video, image-to-video, and audio-conditioned workflows with configurable resolution, duration, and FPS.
P-Image LoRA combines ultra-fast image generation with customizable style adaptation through LoRA weights. Apply community-trained or custom LoRA models from HuggingFace to achieve specific artistic styles, characters, or visual themes while maintaining sub-second generation speed.
P-Image Edit LoRA combines the speed and quality of P-Image Edit with the flexibility of Low-Rank Adaptation (LoRA) fine-tuning. This enables custom style transfer, character consistency, brand-specific aesthetics, and domain-specific edits using pre-trained LoRA weights from HuggingFace. The model maintains sub-second inference while applying custom adaptations, making it ideal for production workflows requiring consistent styling or specialized editing capabilities.
Qwen3-VL is a vision-language instruction-tuned model capable of understanding text and images. It supports streaming, OCR, and rich multimodal conversations.
No models match your search. Try a different keyword or category.
Access enterprise-grade open-source AI models including Llama 3, DeepSeek, Qwen, and more via our high-performance serverless API. Experience low-latency inference on the latest NVIDIA GPUs optimized for production workloads.
"Qubrid AI reduced our document processing time by over 60% and significantly improved retrieval accuracy across our RAG workflows."
Enterprise AI Team
Document Intelligence Platform