Privacy-Preserving On-Prem Deployment

Meta's open-weights multimodal MoE family (Scout & Maverick).

High-performance open MoE LLM rivaling closed frontier models on benchmarks.

Open reasoning model trained with RL, competitive with o1-class systems.

European frontier LLM strong at code, math and multilingual tasks.

Open multilingual model family with hybrid thinking modes.

Enterprise-grade RAG and tool-use model for business workloads.

Small language model punching above its weight on reasoning benchmarks.

Open-weights image generator with strong fine-tuning ecosystem.

State-of-the-art open image model from ex-Stable Diffusion researchers.

Open multilingual speech recognition and translation model.

Code-specialised open model covering 80+ programming languages.

Genome-scale foundation model spanning DNA, RNA and proteins.

Segment Anything for images and video, in real time.

Generative physics platform for robotics simulation and 4D worlds.

Fast single-image to 3D mesh reconstruction model.

OpenAI's first open-weights reasoning models since GPT-2.

Open-weights model family in 1B–27B sizes for on-device & server.

Open vision-language models for fine-tuning.

Open-weights instruct model competitive with much larger LLMs.

Open multimodal model in 11B and 90B sizes.

Self-supervised vision foundation model for image features.

Multilingual speech-to-speech and speech-to-text translation.

Open text-to-music model from AudioCraft.

Fast open-weights small model with strong reasoning.

124B multimodal model with state-of-the-art image understanding.

Sparse mixture-of-experts open model.

Open agentic coding model built with All Hands AI.

Open agentic coding model in the Qwen3 family.

Open vision-language model with strong document understanding.

Open text-to-video and image-to-video model.

Open agentic foundation model from Zhipu's GLM family.

Trillion-parameter open MoE model with strong agentic skills.

Open reasoning model with 1M-token context.

Open MoE model with 389B params from Tencent.

Image editing model with character & style consistency.

Open 17B image generation model topping benchmarks.

Open 13B text-to-video model.

Open-source video generation model.

Real-time open video generation model.

Open ASR model optimised for real-time edge inference.

Open-source CLI pair-programmer using your favorite LLM.

Open-source AI code assistant for VS Code & JetBrains.

Multilingual long-context embedding model.

Open multi-functional, multilingual embedding model.

Open MoE multilingual text embeddings.

Frontier protein language model for biology design.

Open diffusion model for de novo protein design.

Open AlphaFold3-class biomolecular structure prediction model.

Best-in-class medium-range global weather forecasting AI.

Probabilistic AI weather forecasting beating ENS.

Generalist vision-language-action model for robots.

Open foundation model for humanoid robots.

World foundation models for physical AI simulation.

Fully open language model with training data and code released.

Open LLM family from the Technology Innovation Institute.

Compact open model strong in its size class.

Multimodal frontier model with open weights.

NVIDIA's open LLM family for synthetic data and reasoning.

Open enterprise-ready foundation models.

Open enterprise LLM optimised for SQL and coding.

Open MoE LLM tuned for enterprise tasks.

Open software stack for GPU compute and AI on Instinct & Radeon hardware.

Open toolkit to optimize and deploy AI inference across Intel CPUs, GPUs and NPUs.

Open RISC-V based AI accelerator from Jim Keller's team.

Open enterprise LLM family for code, language and time series.

Open MoE LLM by Databricks for enterprise customization.

Open-source vector database with hybrid search and modules.

Segment Anything 3D — reconstructs objects, scenes and human bodies from a single image.

Fully open model flow with training data, checkpoints and recipes for reproducible AI.

Long-context open agentic model from Moonshot, strong on tool use and coding.

Open bilingual frontier model from Zhipu, competitive on coding and reasoning.

Agentic RL vision-language model for tool-integrated visual reasoning.

Toolkit letting AI agents transact on-chain with wallets, USDC and smart contracts.

Google DeepMind's mathematics model tracked by Epoch, focused on geometry.

In recent months, our focus has been on developing a “good” model while optimizing the developer experience.

Recent breakthroughs in large language models (LLMs) have centered around a handful of data-rich languages.

Performing language-conditioned robotic manipulation tasks in unstructured environments is highly demanded for general intelligent robots.

Meta's language model tracked by Epoch, focused on chat.

Visual language models (VLMs) rapidly progressed with the recent success of large language models.

University of California (UC) Berkeley's robotics model tracked by Epoch, focused on robotic manipulation.

After months of efforts, we are pleased to announce the evolution from Qwen1.5 to Qwen2.

High-quality preference datasets are essential for training reward models that can effectively guide large language models (LLMs) in generating high-quality responses aligned with human preferences.

Stanford University's robotics, vision, language model tracked by Epoch, focused on robotic manipulation.

We release the Nemotron-4 340B model family, including Nemotron-4-340B-Base, Nemotron-4- 340B-Instruct, and Nemotron-4-340B-Reward.

We present DeepSeek-Coder-V2, an open-source Mixture-of-Experts (MoE) code language model that achieves performance comparable to GPT4-Turbo in code-specific tasks.

We introduce Cambrian-1, a family of multimodal LLMs (MLLMs) designed with a vision-centric approach.

We're contributing Mathstral to the science community to bolster efforts in advanced mathematical problems requiring complex, multi-step logical reasoning.

Modern artificial intelligence (AI) systems are powered by foundation models.

We present LLaVA-OneVision, a family of open large multimodal models (LMMs) developed by consolidating our insights into data, models, and visual representations in the LLaVA-NeXT blog series.

We present Jamba-1.5, new instruction-tuned large language models based on our Jamba architecture.

DeepSeek's language model tracked by Epoch, focused on language modeling/generation.

In the past three months since Qwen2’s release, numerous developers have built new models on the Qwen2 language models, providing us with valuable feedback.

In the past three months since Qwen2’s release, numerous developers have built new models on the Qwen2 language models, providing us with valuable feedback.

Qwen2.5 is the latest series of Qwen large language models.

Visual data comes in various forms, ranging from small icons of just a few pixels to long videos spanning hours.

China Telecom's language model tracked by Epoch, focused on language modeling/generation.

Meta's multimodal, vision, language model tracked by Epoch, focused on visual question answering.

Tsinghua University's robotics model tracked by Epoch, focused on robotic manipulation.

We introduce Chai-1, a multi-modal foundation model for molecular structure prediction that performs at the state-of-the-art across a variety of tasks relevant to drug discovery.

NVIDIA's vision, language model tracked by Epoch, focused on language modeling/generation.

Visual language models (VLMs) have made significant advances in accuracy in recent years.

We present Infinity, a Bitwise Visual AutoRegressive Modeling capable of generating high-resolution, photorealistic images following language instruction.

This technical report introduces the EXAONE 3.5 instruction-tuned language models, developed and released by LG AI Research.

We study the problem of single-image 3D object reconstruction.

INTELLECT-MATH is a 7B parameter model optimized for mathematical reasoning.

Recently, promising progress has been made by open-source vision-language models (VLMs) in bringing their capabilities closer to those of proprietary frontier models.

Tsinghua University's mathematics model tracked by Epoch, focused on mathematical reasoning.

QwQ is the reasoning model of the Qwen series.

We present EXAONE Deep series, which exhibits superior capabilities in various reasoning tasks, including math and coding benchmarks.

In this report, we introduce ERNIE 4.5, a new family of large-scale multimodal models comprising 10 distinct variants.

Understanding and modeling lighting effects are fundamental tasks in computer vision and graphics.

We present DeepSeek-V3, a strong Mixture-of-Experts (MoE) language model with 671B total parameters with 37B activated for each token.

We’re sharing the first models in the Llama 4 herd, which will enable people to build more personalized multimodal experiences.

We’re sharing the first models in the Llama 4 herd, which will enable people to build more personalized multimodal experiences.

Today, we are excited to announce the release of Qwen3, the latest addition to the Qwen family of large language models.

We introduce our first-generation reasoning models, DeepSeek-R1-Zero and DeepSeek-R1.

In this work, we introduce the Qwen3 Embedding series, a significant advancement over its predecessor, the GTE-Qwen series, in text embedding and reranking capabilities, built upon the Qwen3 foundation models.

LG AI Research's vision, medicine model tracked by Epoch, focused on cancer diagnosis.

This technical report introduces EXAONE 4.0, which integrates a Non-reasoning mode and a Reasoning mode to achieve both the excellent usability of EXAONE 3.5 and the advanced reasoning abilities of EXAONE Deep.

Today, we're announcing Qwen3-Coder, our most agentic code model to date.

Today, we are excited to announce the release of Qwen3, the latest addition to the Qwen family of large language models.

Today, we are excited to announce the release of Qwen3, the latest addition to the Qwen family of large language models.

We present Qwen-Image, an image generation foundation model in the Qwen series that achieves significant advances in complex text rendering and precise image editing.

OpenAI's language model tracked by Epoch, focused on language modeling/generation.

We introduce LongCat-Flash, a 560-billion-parameter Mixture-of-Experts (MoE) language model designed for both computational efficiency and advanced agentic capabilities.

Large language models (LLMs) have evolved into agentic systems capable of autonomous tool use and multi-step reasoning for complex problem-solving.

We present Qwen3-Omni, a single multimodal model that, for the first time, maintains state-of-the-art performance across text, image, audio, and video without any degradation relative to single-modal counterparts.

Ling-1T is the first flagship non-thinking model in the Ling 2.0 series, featuring 1 trillion total parameters with ≈ 50 billion active parameters per token.

Today, we are officially open-sourcing and launching MiniMax M2, a model born for Agents and code.

We present Tongyi DeepResearch, an agentic large language model, which is specifically designed for long-horizon, deep information-seeking research tasks.

Today, we are introducing Kimi K2 Thinking, our best open-source thinking model.

Recent progress in large language models (LLMs) has moved the frontier from puzzle-solving to science-grade reasoning-the kind needed to tackle problems whose answers must stand against nature, not merely fit a rubric.

DeepSeek's language model tracked by Epoch, focused on mathematical reasoning.

We present Nemotron 3 Nano 30B-A3B, a Mixture-of-Experts hybrid MambaTransformer language model.

Zhipu AI's language model tracked by Epoch, focused on language modeling/generation.

MiniMax's language model tracked by Epoch, focused on chat.

NC AI's language model tracked by Epoch, focused on language modeling/generation.

Today, we're announcing Qwen3-Coder-Next, an open-weight language model designed specifically for coding agents and local development.

We introduce Kimi K2.5, an open-source multimodal agentic model designed to advance general agentic intelligence.

We are delighted to announce the official release of Qwen3.5, introducing the open-weight of the first model in the Qwen3.5 series, namely Qwen3.5-397B-A17B.

We present GLM-5, a next-generation foundation model designed to transition the paradigm of vibe coding to agentic engineering.

Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance.

Moonshot AI's language model tracked by Epoch, focused on language modeling/generation.

DeepSeek's language model tracked by Epoch, focused on language modeling/generation.

DeepSeek's language model tracked by Epoch, focused on language modeling/generation.

Liquid AI's second-generation efficient foundation models built on liquid neural networks for on-device use.

LLM-discovered preference optimization algorithm from Sakana's evolutionary research line.

First Indic-first foundation model optimized for 10 Indian languages and English.

Massively multilingual open-weights model covering 23 languages from Cohere For AI.

Open-source aligned LLM family known for steerable, uncensored research use.

Meituan LongCat's open-source audio-driven avatar video model for single- and multi-character human video generation.

Tencent Hunyuan's open-source multilingual translation family for fast, instruction-following translation across 33 languages.

Microsoft's open-source 3.8B text-to-image model focused on efficient training, fast high-res generation, and strong prompt adherence.

Cohere's open-source W4A4-quantized vision-language reasoning model for agentic, multilingual, tool-use enterprise tasks.

NVIDIA's open 14B text-generation LM supporting autoregressive, diffusion-style parallel, and self-speculative decoding.

Meta's audio generation model focused on high-fidelity waveform synthesis and speech-music co-generation.

Sapient's 1B Hierarchical Reasoning Model for compact, structured chain-of-thought text generation.

Stability AI's compact text-to-sound-effects diffusion model optimized for low-latency on-device SFX generation.

Stability AI's compact text-to-music diffusion model tuned for short, license-friendly musical loops and stems.