Llama 4

Top Builders

Explore the top contributors showcasing the highest number of app submissions within our community.

Llama 4

Llama 4 is Meta AI’s newest open-weight model series.
It introduces Mixture-of-Experts (MoE) routing for efficient inference, accepts both text and images natively, and stretches context windows to record-breaking lengths—10 M tokens in the Scout variant. Meta positions Llama 4 as a research-friendly, production-ready alternative to proprietary frontier models, while keeping the code and weights downloadable from its GitHub repos and the official llama.com portal.

General
Release date	5 Apr 2025
Developer	Meta AI
Type	Open-weight multimodal LLM
License	Llama 4 Community License
GitHub	meta-llama/llama-models

Core Features

Mixture-of-Experts architecture – Each query activates a subset of specialised “experts,” yielding higher throughput per FLOP while scaling to trillions of total parameters (TechCrunch).
Native multimodality – Models ingest both text and images without external adapters (The Verge).
Extended context windows – Scout handles up to 10 M tokens; Maverick supports 1 M tokens (llm-stats).
Multilingual training – Optimised across 200+ languages for global deployments (Data Scientist Guide).
Fine-tunable & agent-ready – Models ship with recipes for supervised fine-tuning, LoRA, and RAG inside the Llama Cookbook.

Model Variants

Variant	Active Params	Experts	Total Params	Context Window	Best for
Scout	17 B	16	109 B	10 M tokens	Long-context RAG, document analysis (stats)
Maverick	17 B	128	400 B	1 M tokens	Coding & reasoning tasks, general chat (Oracle Docs)
Behemoth	288 B*	16	~2 T	TBA	High-end STEM, under training (not yet released)

Tools & Resources

Weights & License – Download from the official portal.
Inference & training code – meta-llama/llama-models.
Quick-start recipes – Llama Cookbook.
Stats & benchmarks – Interactive dashboards at llm-stats.com.
Cloud endpoints – Ready-to-use deployments in Oracle OCI Generative AI.

Ecosystem & Integrations

Meta AI assistant now runs Llama 4 across WhatsApp, Messenger, Instagram, and web chat (The Verge).
OCI Generative AI offers managed Scout & Maverick endpoints for enterprise workloads (Oracle Docs).
Community hosting – Providers such as DeepInfra, Groq, and Together price Llama 4 as low as $0.08 / 1 M input tokens (llm-stats).
Research & open-source – Thousands of fine-tuned checkpoints already live on Hugging Face; Meta’s annual LlamaCon (29 Apr 2025) spotlights academic collaborations (TechCrunch).

Llama 4 pushes open-weight LLMs into frontier-model territory—combining trillion-scale capacity with permissive licensing. Start experimenting by cloning the GitHub repo, reading the cookbook, or provisioning a managed endpoint on Oracle OCI.

Edit on GitHub

Meta Llama 4 AI technology Hackathon projects

Discover innovative solutions crafted with Meta Llama 4 AI technology, developed by our community members during our engaging hackathons.

AuraOS

Aura is a sovereign, neuro-symbolic AI operating system designed to make advanced artificial intelligence more transparent, efficient, collaborative, and accountable. Instead of sending every instruction directly to a language model, Aura can locally interpret natural-language intent, organize it into structured semantic and routing representations, identify relevant code or knowledge, and prepare a bounded task before model use is required. Aura combines deterministic routing, a 4,096-primitive lexical system, polysynthetic six-slot intent structures, visual code topology, guarded human-agent workflows, and a Learning Arena where verified experiences can be tested before becoming reusable knowledge. Its Observatory allows users to see how Aura understood an instruction, while the Human Agent Arena supports node-specific dialogue, approval gates, verification, and preserved attempt artifacts—including failed work that may help future refactoring. Aura is built around human authority. Models can propose, explain, investigate, and assist, but they cannot silently expand scope, bypass evidence requirements, or merge changes without approval. The result is an AI architecture focused not only on capability, but on sovereignty, inspectability, learning, and responsible action.

Probably runs on Samsung S25

Hybrid Router is an intelligent AI inference system that dynamically selects the best execution path for each user query instead of relying on a single large language model. The system first analyzes the incoming prompt to identify the task type, estimate its complexity, and determine whether it can be solved deterministically or requires generative AI. For structured tasks such as mathematical calculations, JSON validation, regular expression verification, and date/time operations, the router invokes specialized deterministic tools that produce fast, accurate, and reproducible results without consuming LLM tokens. For more complex natural language and coding tasks, the router attempts local inference using OpenVINO-optimized models running on Intel hardware, reducing latency and API costs. If the local model is unlikely to provide a sufficiently reliable answer or the task exceeds its capabilities, the system automatically falls back to Fireworks AI models for high-quality remote inference. The routing decisions are driven by task classification, confidence estimation, and configurable thresholds, allowing the system to balance accuracy, response time, and operational cost. The architecture is modular, making it easy to add new tools, local models, or routing strategies in the future. The project also includes benchmarking and evaluation components that measure routing accuracy, latency, model utilization, and fallback frequency to continuously improve routing performance. By combining deterministic tools, local inference, and cloud-based language models into a single adaptive pipeline, Hybrid Router delivers efficient, scalable, and cost-aware AI inference while maintaining high response quality across a wide range of tasks.

RouteLM: Adaptive Model Routing Engine

**Hybrid Model Router** is an open-source framework for intelligently routing AI requests across multiple large language models (LLMs). Instead of sending every request to a single model, it evaluates configurable routing strategies to choose the most suitable model for each task. The router is designed to optimize factors such as response quality, inference cost, latency, context length, and model availability. It provides a unified interface for integrating multiple AI providers, making it easier to build applications that balance performance and efficiency without changing application code. ### Features * Intelligent request routing across multiple AI models. * Configurable routing policies based on cost, latency, capability, or custom rules. * Unified API for interacting with different model providers. * Extensible architecture for adding new providers and routing strategies. * Modular design suitable for production deployments and experimentation. * Lightweight and easy to integrate into existing AI applications. Whether you're building chatbots, AI agents, coding assistants, or enterprise applications, Hybrid Model Router helps maximize model performance while minimizing operational costs by automatically selecting the right model for every request.

Frugal Route

FrugalRoute is a token-efficient LLM routing agent built for the AMD Developer Hackathon ACT II, Track 1. Instead of sending every request to an expensive remote model, it first tries a free local model served on an AMD MI300X GPU via Ollama, giving that model access to real tools (calculator, web search) to answer directly when possible. A calibrated confidence gate — fit with Platt/Isotonic scaling and an AutoMix POMDP belief model — decides, per request, whether the local answer is trustworthy enough to keep or whether to escalate to a paid Fireworks-hosted model. A semantic cache and prompt compression layer further cut redundant remote calls. The result: most traffic gets resolved for free, with paid tokens reserved only for genuinely hard cases, all while an enforced accuracy floor keeps output quality high.

ISEEKAIGO

Most AI companions today are built around static prompts layered on top of large language models. They can generate impressive responses, but they rarely evolve beyond the conversation in front of them. ISEEKAIGO is building the next generation of AI companionship by combining generative AI with persistent memory, human-inspired needs, autonomous behavior, and evolving cognition. Instead of simply responding, our AI learns from experience, self-learning, adapts over time, and develops its own behavioral context, making every interaction more personal and meaningful. Our vision extends beyond creating a better chatbot. We are building a foundation for Human-like AI that can interact with people over long periods while becoming increasingly capable through experience. By breaking this vision into scalable milestones, we aim to transform AI companionship into a platform for future autonomous AI applications such as building physical-AI core. We believe this project represents a meaningful step toward the future of Human-like AI, combining ambitious research with the ability to ship real products and bring advanced AI experiences to millions of users.

MEIA-LAB (Multi-Agent Earnings Intelligence calls)

MEIA-LAB (Multi-Agent Earnings Intelligence & Alignment Lab) is an end-to-end AI platform that automates the verification of corporate earnings call claims against official SEC filings in real-time. Every quarter, public companies present their financial performance through earnings calls and slide decks, but analysts must manually cross-reference hundreds of pages of SEC disclosures to separate fact from corporate spin—a process that is slow, error-prone, and costly. MEIA-LAB solves this by deploying a team of specialized AI agents orchestrated through LangChain and LangGraph. The ASR Agent uses Whisper to transcribe earnings call audio with high accuracy. The Vision Agent performs layout-aware OCR on presentation slides using PyMuPDF and multimodal LLMs. The Filing Agent retrieves the latest 10-Q and 8-K filings directly from SEC EDGAR and extracts structured financial data. Finally, the Orchestrator cross-examines all sources, scores each management claim for consistency, flags risk discrepancies, and generates source-cited analyst briefs. The platform features a React-based dashboard that surfaces a real-time Consistency Score, a verified claims panel with direct evidence citations, a risk monitor highlighting guidance gaps, and an Evidence Room linking every finding to exact filing pages and transcript lines. All heavy inference workloads including Whisper transcription and high-throughput vector embedding generation for RAG-based filing search via ChromaDB are accelerated on AMD Developer Cloud GPUs using ROCm, delivering massive speedups over CPU execution. MEIA-LAB cuts institutional-grade earnings research from hours to under a minute, helping hedge funds, equity research desks, and compliance teams invest with certainty.