Trusted by 2,400+ Researchers & Teams

The Open Marketplace
for AI Research
& Foundation Models

Discover, evaluate, and deploy cutting-edge LLMs, Mixture-of-Experts architectures, NLP systems, and self-supervised learning models. From open research to production-grade deployments.

🔥 Trending LLM 7B–405B Mixture of Experts NLP · NER · QA Self-Supervised Multimodal Free Models
🧠
PGR-MoE-72B
puregamma / mixtral-arch
Sparse mixture-of-experts LLM with 8 active experts per token. State-of-the-art reasoning & code generation.
MoE 72B ⬇ 48.2K ★ 1.2K
💬
PGR-BERT-NLP-xl
puregamma / nlp-core
Bidirectional encoder for NER, QA, classification. Fine-tuned on 140 languages.
NLP 1.5B ⬇ 31.5K
🔬
PGR-SSL-ViT-Large
puregamma / ssl-lab
Self-supervised vision transformer pretrained with DINO v2 + MAE hybrid objective.
SSL 307M ⬇ 19.3K
0
Models Listed
0
Researchers
0
Monthly Downloads
0
Research Papers
0
Languages Supported
LLaMA 3.1 · 405B Mistral MoE · 8×7B BERT · RoBERTa · DeBERTa GPT-4 Class Models DINO v2 · MAE · CLIP Reinforcement Learning Chain-of-Thought Reasoning Multimodal Vision-Language Code Generation · StarCoder Autonomous Agents · ReAct LLaMA 3.1 · 405B Mistral MoE · 8×7B BERT · RoBERTa · DeBERTa GPT-4 Class Models DINO v2 · MAE · CLIP Reinforcement Learning Chain-of-Thought Reasoning Multimodal Vision-Language Code Generation · StarCoder Autonomous Agents · ReAct
Model Hub

Featured Models

Browse all models →
🦙
FREELLM
puregamma / llama-series
PGR-LLaMA-3.1-70B-Instruct
Fine-tuned instruction-following model based on LLaMA 3.1 architecture. Excels at reasoning, code synthesis, and multilingual generation with 128K context window.
Text GenerationInstruction FollowingCodeMultilingual
PROMoE
puregamma / moe-lab
PGR-MoE-Mixtral-8x22B
Sparse mixture-of-experts with 8 active experts per token from 56 total. Outperforms dense 70B models at a fraction of inference cost. Optimized for long-context tasks.
MoELong ContextEfficient Inference
💬
FREENLP
puregamma / nlp-core
PGR-DeBERTa-v3-NLP-XL
State-of-the-art encoder for Named Entity Recognition, Question Answering, text classification and relation extraction. Fine-tuned on 2.4B tokens across 24 NLP benchmarks.
NERQAClassificationRE
🔬
FREESSL
puregamma / ssl-lab
PGR-DINO-ViT-H-SSL
Vision transformer pretrained with self-supervised DINO v2 objective. Learns rich patch-level semantic features without labels. Achieves 86.1% ImageNet top-1 with linear probing.
VisionSelf-SupervisedFeature Extraction
🎯
PRORL
puregamma / autonomous-ai
PGR-RLHF-Agent-7B
Autonomous reasoning agent trained with PPO + RLHF. Capable of multi-step tool use, web browsing, code execution, and long-horizon planning tasks with ReAct framework.
RLHFAgentTool UseReAct
🌐
FREELLM
puregamma / multilingual
PGR-Qwen-2.5-14B-zh-en
Bilingual Chinese-English LLM with strong performance on Asian language benchmarks. 48K context window, function calling, and structured output capabilities.
ChineseEnglishFunction Calling
Explore

Browse by Research Area

🧠
Large Language Models
4,820 models
Mixture of Experts
1,240 models
💬
NLP & Language
3,650 models
🔬
Self-Supervised Learning
980 models
🤖
Autonomous / RL Agents
710 models
Open LLM Leaderboard

Model Rankings

Full leaderboard →
#
Model
Score
Performance
Params
License
1
🦙
PGR-LLaMA-3.1-405B
puregamma
89.4
405B
Enterprise
2
PGR-MoE-Mixtral-8x22B
puregamma
87.1
141B
PRO
3
🦙
PGR-LLaMA-3.1-70B-Instruct
puregamma
84.6
70B
FREE
4
🌐
PGR-Qwen-2.5-72B
puregamma
83.2
72B
FREE
5
🎯
PGR-RLHF-Agent-70B
puregamma
81.8
70B
PRO
Research Hub

Latest Papers & Datasets

View all research →
PAPER Mar 2025
Scaling Sparse Mixture-of-Experts Beyond 1T Parameters: Routing Strategies and Expert Specialization
We investigate routing dynamics and expert specialization in MoE architectures scaled beyond 1 trillion parameters. Our analysis reveals emergent specialization patterns and proposes a novel auxiliary loss that improves expert utilization by 31%.
Chen, L. · Park, S. · Williams, R. · et al. · Pure Gamma Research Lab
DATASET Feb 2025
PGR-Instruct-2M: High-Quality Instruction Tuning Dataset for Multilingual LLM Alignment
A curated 2-million sample instruction dataset spanning 48 languages, with human-verified quality scores. Includes chain-of-thought reasoning traces, multi-turn dialogues, and tool-use demonstrations for RLHF.
Pure Gamma Research · Data Curation Team · 2025
BENCHMARK Jan 2025
PGR-Eval: A Comprehensive Evaluation Suite for Self-Supervised Representation Learning
Multi-task evaluation framework covering 42 downstream tasks for assessing self-supervised representations. Tests feature quality, sample efficiency, and transfer learning across vision, language, and multimodal settings.
SSL Research Group · Pure Gamma Research · NeurIPS 2025
TOOL Apr 2025
PGR-FineTune: One-Click PEFT Fine-tuning Infrastructure for LLMs up to 405B
Production-ready fine-tuning toolkit supporting LoRA, QLoRA, DoRA, and full fine-tuning. Automatic hyperparameter selection, gradient checkpointing, and multi-node distributed training on consumer and enterprise hardware.
MLOps Team · Pure Gamma Research Infrastructure · 2025
Platform

Deploy Models in Minutes

1
Discover & Evaluate
Browse 12,400+ models filtered by architecture, task, parameters, license, and benchmark scores. Read model cards, review research papers, and compare on our open leaderboard.
2
Download or Fine-tune
Pull models directly via our Python SDK, compatible with Hugging Face Transformers, vLLM, and GGUF formats. Use PGR-FineTune for one-click LoRA / QLoRA fine-tuning on your data.
3
Publish Your Research
Upload your model, dataset, or paper. Get a DOI, versioned model card, and automatic benchmark evaluation. Earn from commercial licenses or release under Apache / MIT open license.
4
Deploy to Production
One-click serverless inference endpoints with auto-scaling, quantization (INT4/INT8/FP16), and OpenAI-compatible API. Pay per token or reserve dedicated GPU capacity.
pgr-sdk · Python 3.11
# Install PGR SDK # pip install pgresearch from pgresearch import ModelHub # Load a model directly model = ModelHub.load( "puregamma/PGR-LLaMA-3.1-70B", quantize="int4", device="cuda" ) # Run inference response = model.generate( "Explain MoE routing in detail:", max_tokens=512, temperature=0.7 ) print(response.text)
Pricing

Choose Your Research Plan

Access open-source models for free. Unlock premium models, fine-tuning, and inference for teams.

Researcher
Free
Forever free · Open source models
  • 4,820+ free LLM models
  • Community NLP benchmarks
  • Public datasets & papers
  • Model card hosting (10 models)
  • Python SDK access
  • Inference API
  • Fine-tuning service
  • Premium PRO models
Get Started Free
Enterprise
Custom
Team & org pricing · SLA guaranteed
  • Everything in Pro
  • Unlimited inference tokens
  • Dedicated GPU clusters
  • On-premise deployment
  • Custom model fine-tuning
  • Private marketplace access
  • 99.9% SLA + 24/7 support
  • IP & commercial licensing
Contact Sales
Join the Platform

Build the Future of
AI Research

Publish your models and research. Collaborate with 2,400+ AI researchers. Access production inference infrastructure for LLMs, MoE, NLP, and beyond.