AI inference

Models 16

• Gen 3
  DiffusionGemma

  By Google

  DiffusionGemma is Google’s open experimental 26B MoE text-diffusion model for faster local text generation, parallel editing, code infilling, and non-linear text workflows.

  💬Chatting🔄Language model optimization🧠AI inference
  NewMultimodal

  Released 1mo ago
• Gen 3
  SubQ 1M Preview

  By Subquadratic

  SubQ 1M-Preview is Subquadratic’s first large language model and the first fully subquadratic LLM, built so compute grows linearly with context length rather than quadratically. It is designed for massive-context reasoning, coding, search, and full-corpus processing, making million-token windows more practical, faster, and cheaper than standard transformer-based approaches

  📚Large Language Models🧠AI inference🦾Ai research
  NewMultimodal

  Released 2mo ago
• Gen 3 DeepSeek
  DeepSeek V4 Flash

  By DeepSeek

  DeepSeek-V4-Flash is DeepSeek’s faster, smaller, and more economical V4 model for efficient large-scale use. It has 284B total parameters with 13B active parameters, supports a 1M token context window, and is positioned as a high-speed, cost-effective model whose reasoning comes close to V4-Pro.

  📚Large Language Models💬Chatting💻Conversational coding🧠AI inference
  NewMultimodal

  Released 2mo ago
• Gen 3 DeepSeek
  Deepseek V4 Pro

  By DeepSeek

  DeepSeek-V4-Pro is DeepSeek’s new flagship open-source model for top-end reasoning, coding, world knowledge, and agentic work. It uses a MoE architecture with 1.6T total parameters and 49B active parameters, supports a 1M token context window, and is positioned as rivaling leading closed-source models while leading current open models on several major capabilities.

  📚Large Language Models💬Chatting💻Conversational coding🧠AI inference
  NewMultimodal

  Released 2mo ago
• Gen 3 Gemma
  Gemma 4 31B IT NVFP4

  By NVIDIA

  Gemma-4-31B-IT-NVFP4 is NVIDIA’s inference-optimized NVFP4 quantized version of Gemma 4 31B IT. It is a commercial-ready multimodal model for text, image, and video understanding with text output, built for reasoning, coding, chat, and agentic workflows while preserving the original model’s long 256K context window.

  🔄Language model optimization🔍Image interpretation🧠AI inference
  Multimodal

  Released 3mo ago
• Gen 7 Trinity
  Trinity Large Thinking

  By Arcee AI

  Trinity-Large-Thinking is Arcee AI’s open-source reasoning model for long-horizon agents and multi-turn tool use. It is built to improve context coherence, instruction following, and stability across extended agent loops, and is released under Apache 2.0 for developers and enterprises that want open weights they can inspect, host, and post-train.

  💬Chatting🔍Code reviews💻Conversational coding🧠AI inference
  Text

  Released 3mo ago
• Gen 7
  1 Bit Bonsai 1.7B

  By PrismML

  1-bit Bonsai 1.7B is PrismML’s smallest Bonsai model, built for extremely fast on-device inference. With a memory footprint of just 0.24GB, PrismML says it can reach 130 tokens per second on an iPhone 17 Pro Max, making it a lightweight edge model aimed at mobile and highly power-constrained applications.

  🧠AI inference🤖Ai deployment management💰Ai cost management
  Text

  Released 3mo ago
• Gen 7
  1 Bit Bonsai 4B

  By PrismML

  1-bit Bonsai 4B is PrismML’s mid-size ultra-efficient model designed for fast local inference. It uses just 0.57GB of memory and is positioned as a strong balance of speed, accuracy, and energy efficiency, with PrismML reporting throughput of 132 tokens per second on an M4 Pro for workloads that need both responsiveness and capable language performance.

  🧠AI inference🤖Ai deployment management💰Ai cost management
  Text

  Released 3mo ago
• Gen 7
  1 Bit Bonsai 8B

  By PrismML

  1-bit Bonsai 8B is PrismML’s flagship ultra-efficient language model, built with 1-bit weights for edge and real-time use. It needs only 1.15GB of memory and is aimed at robotics, real-time agents, and on-device computing, with PrismML claiming a 14x smaller footprint, 8x faster runtime, and 5x better energy efficiency than standard full-precision 8B models.

  🧠AI inference🤖Ai deployment management💰Ai cost management
  Text

  Released 3mo ago
• Gen 7
  LFM2.5 350M

  By Liquid AI

  LFM2.5-350M is Liquid AI’s compact 350M-parameter language model optimized for fast inference, tool use, data extraction, and structured outputs. It is designed to run across cloud GPUs, CPUs, and edge devices, and is positioned for lightweight agentic workflows and large-scale data processing rather than math, coding, or creative writing.

  📚Large Language Models🔍Data extraction🔄Language model optimization🧠AI inference
  Text

  Released 3mo ago
• Gen 7
  Mamba-3

  By Together AI

  Mamba-3 is a new state space model (SSM) architecture designed with inference efficiency as the primary goal, improving prefill and decode latency across sequence lengths.

  🔨LLM development🎓LLM training🔄Language model optimization🧠AI inference
  Text

  Released 4mo ago
• Gen 7
  Sequential Hidden Decoding 8B n8

  By Tencent

  Sequential-Hidden-Decoding-8B-n8 is Tencent’s base language model variant that extends effective sequence length by 8x using multiple embedding passes while keeping the same Transformer backbone. Built on Qwen3-8B-Base, it is aimed at benchmarking, text completion, and fine-tuning research, with hidden latent steps acting like implicit reasoning before each predicted token.

  📚Large Language Models🤖Ai research assistance🔄Language model optimization🧠AI inference
  Text

  Released 4mo ago
• Gen 3
  FlashDrive

  By Z Lab

  FlashDrive is Z Lab’s real-time inference framework for vision-language-action autonomous driving models. It is designed to make reasoning-based end-to-end driving practical by accelerating all major inference stages, reducing latency from 716 ms to 159 ms on Alpamayo 1.5 while keeping accuracy nearly unchanged.

  🧠AI inference🗃️Model utilization🚗Autonomous driving
  Multimodal

  Released 4mo ago
• Gen 7
  llm scaler vllm 0.14.0 b8

  By Intel

  llm-scaler-vllm 0.14.0-b8 is Intel’s Docker-based packaging of vLLM for running LLM inference on Intel Arc and Battlemage GPUs.

  🧠AI inference🚀Performance optimization🤖Ai deployment management🔍Ai system monitoring
  Text

  Released 4mo ago
• Gen 3
  Alpie Core

  By 169Pi

  Alpie-Core is 169Pi’s first-generation 4-bit quantized reasoning model, built on the DeepSeek-32B family and optimized for efficient reasoning, mathematics, coding, and science tasks. It combines QLoRA, LoRA, and synthetic dataset distillation, supports a 65K-token context, and is designed to deliver strong reasoning performance with far lower memory and compute costs.

  💬Chatting💻Conversational coding🧠AI inference
  Text

  Released 9mo ago
• Gen 7
  Marco-o1

  By Alibaba

  Marco-o1 is Alibaba International Digital Commerce’s open reasoning model line aimed at real-world problem solving, not just standard-answer tasks. It is designed for mathematics, physics, coding, planning, and open-ended solutions, with an emphasis on broader agentic use, including planning ability and function-calling readiness.

  💬Chatting🌐Text translation💻Conversational coding🧠AI inference
  Text

  Released 1y ago