Ollama
Local LLM Runners
- The default way to run LLMs locally
- One command to run Llama, Qwen, DeepSeek…
- OpenAI-compatible local API
- Free and open-source
AI Models & Local Execution (67)
All 67 ai models & local execution tools in the Flocci catalog, 64 of them free or freemium. Pricing tier and real capabilities on every entry — no ads, no sponsored placement.
AllLocal LLM RunnersModel Hosting & Inference APIsNotable Open ModelsSpeech & Transcription EnginesFine-Tuning & Training FrameworksLLM Eval & ObservabilityLeaderboards & BenchmarksRobotics & Embodied AI
free
LM Studio
Local LLM Runners
- Polished desktop GUI for local models
- Discover, download and chat with GGUF models
- Local server with OpenAI-compatible API
- Free for personal use
Jan
Local LLM Runners
- Open-source, offline ChatGPT alternative
- No-config, privacy-first GUI
- Runs fully on your device
- Free forever
GPT4All
Local LLM Runners
- Private local chat with your documents
- Runs on modest laptops
- LocalDocs RAG built in
- Free and open-source
Open WebUI
Local LLM Runners
- Self-hosted ChatGPT-style interface
- Works with Ollama and OpenAI-compatible APIs
- RAG, tools and multi-user
- Free and open-source
AnythingLLM
Local LLM Runners
- All-in-one local RAG and agents app
- Chat with your docs privately
- Works with local or cloud models
- Free and open-source
llama.cpp
Local LLM Runners
- Pure C/C++ inference engine with minimal dependencies, runs on CPU-only hardware
- Powers the GGUF format used by Ollama, LM Studio and most local runners under the hood
- Broad quantization support (2-bit to 8-bit) for running large models on consumer hardware
vLLM
Local LLM Runners
- PagedAttention memory management for high-throughput multi-request GPU serving
- Supports 200+ model architectures with an OpenAI-compatible API server
- Originated at UC Berkeley Sky Computing Lab, now the de facto standard for self-hosted production LLM serving
SGLang
Local LLM Runners
- RadixAttention for automatic prefix-cache reuse across requests
- Day-0 support for newly released open models (e.g. Kimi K3) among the fastest of any runner
- Scales from single GPU to distributed multi-node clusters with speculative decoding
KoboldCpp
Local LLM Runners
- Ships as a single portable executable with a bundled web UI (KoboldAI Lite)
- Generates text, image, video and speech from one local app
- Runs on CPU or GPU with no installation required
MLX-LM (Apple)
Local LLM Runners
- Purpose-built for Apple Silicon's unified memory architecture, faster than llama.cpp for many models on Mac
- Built-in LoRA/QLoRA fine-tuning support, not just inference
- Direct Hugging Face Hub integration for pulling and pushing quantized models
Msty
Local LLM Runners
- Nexus module: unified gateway for managing model connections, credentials and usage policy across providers
- Governance features (SSO, audit logs, zero telemetry) aimed at private/enterprise use
- Combines local model chat, agents (Go), and knowledge packages (Stack) in one app
Hugging Face
Model Hosting & Inference APIs
- The hub for open models and datasets
- Spaces to host and demo apps
- Inference endpoints and providers
- Free accounts and hosting
Replicate
Model Hosting & Inference APIs
- Run open models with one API call
- Thousands of community models
- Deploy your own with Cog
- Pay-per-second, free trial credits
Groq
Model Hosting & Inference APIs
- Ultra-fast inference on custom LPUs
- Sub-second responses for open models
- OpenAI-compatible API
- Generous free tier
Together AI
Model Hosting & Inference APIs
- Fast, cheap open-model inference
- 200+ models via one API
- Fine-tuning and dedicated endpoints
- Free starter credits
OpenRouter
Model Hosting & Inference APIs
- One API for 400+ models across providers
- Automatic fallbacks and price routing
- Some models free to use
- Unified billing
Fireworks AI
Model Hosting & Inference APIs
- High-performance open-model serving
- Fine-tuning and function calling
- Vision and audio models
- Free trial credits
Cerebras Inference
Model Hosting & Inference APIs
- Runs open models on Cerebras Wafer-Scale Engine hardware at dramatically higher tokens/sec than GPU-based APIs
- New accounts get free credits across all Cerebras-hosted models
- Popular backend choice for latency-sensitive agentic and coding tools
SambaNova Cloud
Model Hosting & Inference APIs
- Hosts open models (DeepSeek, MiniMax, Llama) on SambaNova's own RDU chips for high-speed inference
- Per-token pricing with no platform subscription required
- Competes directly with Cerebras/Groq on raw tokens/sec for open-weight models
DeepInfra
Model Hosting & Inference APIs
- Very low per-token pricing across a large catalog of open models (text, image, audio)
- Flex scheduling tier at 0.8x price for non-production/batch workloads
- No subscription; usage-tier system auto-scales billing thresholds
Novita AI
Model Hosting & Inference APIs
- Single API surface for 200+ LLM/image/video/audio models plus on-demand GPU and bare-metal instances
- Agent Sandbox for secure agent runtimes alongside model hosting
- 'Free to start' onboarding credits across model and GPU products
Nebius Token Factory
Model Hosting & Inference APIs
- European GPU-cloud-backed inference API hosting major open models (Llama, Qwen, DeepSeek, etc.)
- Rebranded in 2026 from 'Nebius AI Studio' to 'Nebius Token Factory'
- Backed by Nebius's own large-scale GPU cloud infrastructure rather than reselling capacity
Cloudflare Workers AI
Model Hosting & Inference APIs
- 10,000 free 'Neurons' of inference per day on every account, resetting daily
- Runs open models (Llama, Mistral, Whisper, etc.) on Cloudflare's global edge network for low-latency serverless inference
- Bundled with Workers/Pages for building full serverless AI apps without managing GPUs
Baseten
Model Hosting & Inference APIs
- Basic plan is pay-as-you-go with free starter credits for new accounts
- Deploys custom/fine-tuned models, not just a fixed model catalog, with per-second GPU billing
- Hosts current open models (DeepSeek, GPT-OSS, Kimi K3) as ready-to-call APIs alongside custom deployment
Google AI Studio
Model Hosting & Inference APIs
- Free-tier API access to Gemini Flash-family models with generous token allowances, no credit card required to start
- Browser-based playground for prompt design, multimodal testing, and one-click API key generation
- Batch API offers 50% cost reduction once on the paid tier
Llama (Meta)
Notable Open Models
- Meta's flagship open-weight family (Llama 4)
- Multimodal and long-context variants
- Run locally or via any host
- Free, permissive license
Qwen (Alibaba)
Notable Open Models
- State-of-the-art open MoE models (Qwen3.x)
- Top-tier coding and multilingual skills
- Runs on a single high-RAM Mac
- Free open weights
DeepSeek Models
Notable Open Models
- Open V3-series and R1 reasoning models
- Efficient MoE, low-cost inference
- Strong math and code
- Free open weights
Gemma (Google)
Notable Open Models
- Google's lightweight open models (Gemma 4)
- Runs in as little as 16GB RAM
- On-device and multimodal variants
- Free and open
GPT-OSS (OpenAI)
Notable Open Models
- OpenAI's open-weight reasoning models
- Run locally or self-host
- Configurable reasoning effort
- Free, permissive license
Pleias
Notable Open Models
- Fully open-weight, open-dataset small language models (350M-3B) explicitly built for EU AI Act compliance
- Trained exclusively on the 2-trillion-token open Common Corpus, not scraped/licensed text
- Purpose-built RAG models (Pleias-RAG-350M/1B) with built-in citation tracking, Apache 2.0 licensed
Mistral Open Models (Magistral, Devstral, Voxtral)
Notable Open Models
- Devstral: open coding-agent model tuned for SWE-bench-style tasks
- Magistral: open reasoning model with visible chain-of-thought
- Voxtral: open-weights speech model for transcription and voice agents
GLM (Z.ai)
Notable Open Models
- GLM-4.5-Flash and GLM-4.7-Flash text models are free via API
- Open-weight GLM checkpoints on Hugging Face (MIT-style license for GLM-4.5-Air)
- Strong agentic/coding benchmark results rivaling DeepSeek and Qwen
IBM Granite
Notable Open Models
- Apache 2.0 licensed enterprise-grade model family
- Hybrid Mamba/transformer architecture (Granite 4) for lower memory footprint
- Built-in focus on governance, provenance and enterprise compliance
OLMo (Ai2)
Notable Open Models
- Only major model family shipping full pretraining data (Dolma), code and intermediate checkpoints, not just weights
- Olmo 3 ships Base, Think, and Instruct variants at 7B/32B
- OlmoCore training framework and Open Instruct pipeline are also open source
Hunyuan (Tencent)
Notable Open Models
- Open-weight large text MoE models (Hunyuan A13B/Hy3 series) with permissive commercial license
- Hunyuan3D line for fast image-to-3D generation (under 1 second)
- Hy-MT translation models covering dozens of language pairs
Falcon (TII)
Notable Open Models
- Falcon H1 hybrid transformer-Mamba architecture for efficient long-context inference
- Falcon Mamba: pure state-space model variant, no attention layers
- Free commercial use under TII's permissive open license
EXAONE (LG AI Research)
Notable Open Models
- EXAONE 4.5: LG's first open-weight vision-language model for industrial use cases
- 33B flagship plus FP8/AWQ quantized variants for efficient local deployment
- Backed by LG's enterprise/manufacturing domain tuning
MiniMax M2 (open weights)
Notable Open Models
- Open-weight MoE text/reasoning model line distinct from MiniMax's Hailuo video product already listed
- Actively iterated through 2026 (M2.1, M2.5, M2.7 checkpoints)
- Competitive on agentic coding and tool-use benchmarks among open Chinese models
Phi (Microsoft)
Notable Open Models
- MIT-licensed small language models optimized for on-device and edge deployment
- Phi-4-mini and Phi-4-multimodal handle text, vision and audio in one small model
- Distributed via Azure AI Foundry, Hugging Face, and Ollama
Nemotron (NVIDIA)
Notable Open Models
- NVIDIA's open post-trained/distilled model family built on Llama and NVIDIA's own architectures
- Includes top-ranked open embedding models (Nemotron Embed, Llama-Embed-Nemotron)
- OpenReasoning-Nemotron: distilled reasoning models for local/edge inference
OpenAI Whisper
Speech & Transcription Engines
- Open-source speech-to-text, run locally free
- 99+ languages and translation
- Robust to accents and noise
- Powers countless apps
Deepgram
Speech & Transcription Engines
- Fast, accurate speech-to-text API
- Real-time streaming and diarization
- Voice-agent building blocks
- Free credits to start
AssemblyAI
Speech & Transcription Engines
- Speech-to-text with audio intelligence
- Summaries, topics and sentiment
- Speaker diarization
- Free tier for developers
NVIDIA Parakeet
Speech & Transcription Engines
- Tops the Hugging Face Open ASR Leaderboard with 6.05% average WER at 0.6B params
- CC-BY-4.0 license, free for commercial and non-commercial use
- RTFx ~3,386 — extremely fast inference relative to accuracy
Mistral Voxtral
Speech & Transcription Engines
- Apache 2.0 open-weight models (24B and 3B) downloadable from Hugging Face
- Built-in Q&A/summarization over audio, not just raw transcription
- Function-calling directly from voice input for agent workflows
Speechmatics
Speech & Transcription Engines
- Free starting credit with no card required, 56+ language support
- Model-training discount program cuts costs 33% for data-sharing customers
- On-premises/VPC deployment for privacy-sensitive enterprise use
Soniox
Speech & Transcription Engines
- Real-time speech-to-text priced well under typical competitor rates
- Combines transcription + translation in one streaming API call
- Diarization and speech translation bundled at no extra cost
ElevenLabs Scribe
Speech & Transcription Engines
- Scribe v2 Realtime transcribes in under 150ms latency
- Dynamic audio-event tagging (laughter, footsteps) beyond plain words
- Keyterm prompting locks in up to 1,000 specified terms for accuracy
WhisperX
Speech & Transcription Engines
- 70x real-time transcription speed via batched inference on top of Whisper
- Adds wav2vec2 forced alignment for accurate word-level timestamps Whisper lacks natively
- Built-in speaker diarization and VAD preprocessing to cut hallucinations
Unsloth
Fine-Tuning & Training Frameworks
- Fine-tunes LLMs, diffusion, TTS and embedding models 2x faster with ~70% less VRAM than standard Hugging Face training
- Free Google Colab notebooks let anyone fine-tune open models like Llama/Qwen/GLM on a free T4 GPU
- Supports LoRA, QLoRA, full fine-tuning, GRPO and DPO reinforcement/preference tuning in one library
Axolotl
Fine-Tuning & Training Frameworks
- Config-file-driven fine-tuning workflow (YAML) covering LoRA, QLoRA, full fine-tuning, preference tuning and RL
- Broad multi-model and multimodal training support with GPU-efficiency optimizations built in
- Popular choice for reproducible post-training recipes shared across the open-model community
LLaMA-Factory
Fine-Tuning & Training Frameworks
- Zero-code Web UI (LLaMA Board) for fine-tuning 100+ open models without writing training scripts
- Supports full-tuning, LoRA, 2/3/4/5/6/8-bit QLoRA, DPO, PPO, GaLore and PiSSA in one framework
- Used internally by Amazon, NVIDIA and Aliyun for open-model post-training
Langfuse
LLM Eval & Observability
- Fully open-source and self-hostable for free via Docker Compose/Kubernetes, not just a hosted SaaS
- Hobby cloud plan is free with no credit card, 50k observability units/month
- Combines tracing, prompt management, evals and datasets in one open platform
Braintrust
LLM Eval & Observability
- Free Starter plan ships model credits and scored evals with no card required
- Unlimited users/projects/datasets/playgrounds even on the free tier
- Built-in playground for side-by-side prompt/model comparison against production traces
Promptfoo
LLM Eval & Observability
- Open-source CLI/library for evaluating prompts, models and RAG pipelines side by side, runs in CI
- Automated red-teaming to surface prompt injection, jailbreak and data-leak vulnerabilities
- 300,000+ user community with an enterprise tier for teams needing hosted guardrails
W&B Weave
LLM Eval & Observability
- Agent-native tracing model with sessions, steps, tools and sub-agents as first-class concepts (not generic spans)
- Pre-built safety/quality scorers for toxicity, bias, PII and hallucination detection out of the box
- Built on Weights & Biases' existing ML-experiment infrastructure, useful for teams already on W&B
Arize Phoenix
LLM Eval & Observability
- Fully open-source, self-hostable with zero setup via `uvx` (pip/conda also supported)
- Combines tracing, evals, datasets, experiments and prompt playground in one local-first tool
- AI engineering agent (PXI) built in for automated troubleshooting of traces
LMArena (Arena.ai)
Leaderboards & Benchmarks
- Crowd-voted 'Battle Mode' blind head-to-head comparisons across text, image, and coding/agent models
- The most-cited human-preference leaderboard in the industry, referenced in nearly every model launch announcement
- Rebranded from LMArena to Arena.ai in 2026, now also covering agent and design-to-code leaderboards
Artificial Analysis
Leaderboards & Benchmarks
- Independent cross-provider benchmarking of intelligence, speed and price for the same model across different hosting APIs
- Artificial Analysis Intelligence Index aggregates multiple benchmarks into one comparable score
- Covers coding agents, image/video, speech and vertical (finance/legal/health) evaluations, not just chat
SWE-bench
Leaderboards & Benchmarks
- The standard benchmark for evaluating autonomous coding agents on real GitHub issue resolution
- Multiple official variants (Verified, Multimodal, Multilingual, Lite) for different evaluation needs
- Companion tooling (SWE-agent, mini-SWE-agent) for running the benchmark yourself, fully open
Design Arena
Leaderboards & Benchmarks
- Crowdsourced Bradley-Terry (Elo-style) ranking of AI models specifically on design/aesthetic quality, not raw capability
- Separate leaderboards for Website, UI Component, Game Dev, Data Viz, 3D, Image, Video and Logo generation
- Real-time rankings from user votes across 140+ countries rather than a fixed static benchmark
Epoch AI
Leaderboards & Benchmarks
- Independent, non-vendor research institute tracking AI compute, training cost and capability trends over time
- Maintains FrontierMath and the Epoch Capabilities Index, benchmarks designed to resist saturation/gaming
- Interactive Data Explorers covering 3,200+ tracked models, data centers, chips and companies, free to browse
LeRobot (Hugging Face)
Robotics & Embodied AI
- Open PyTorch library for training real-world robot control policies (ACT, Diffusion Policy, VLA models)
- Direct integration with Hugging Face Hub for sharing datasets and pretrained robot policies
- Works with low-cost hobbyist robot arms (SO-100/101), not just simulation
NVIDIA Isaac GR00T
Robotics & Embodied AI
- Open, Apache-2.0-licensed vision-language-action foundation model for generalist humanoid robots (GR00T N1.7)
- Opened to all developers worldwide Apr 2026 (previously partner-only), running on DGX Cloud
- Full pipeline: pretrained model + simulation fine-tuning + deployment tools, not just a research checkpoint
Genesis
Robotics & Embodied AI
- Unified multi-physics simulation engine plus photorealistic renderer for robotics/embodied-AI training, fully open source
- 30,000+ GitHub stars, actively updated as recently as Aug 2026
- Pythonic API designed to be dramatically faster than prior simulators for RL/robot-policy training
AI Models & Local Execution — straight answers
What ai models & local execution tools does Flocci list?
Flocci AI Tools lists 67 ai models & local execution tools, ordered free tiers first: AnythingLLM, Arize Phoenix, Axolotl, DeepSeek Models and Design Arena. 64 of the 67 are free or have a free tier. Every entry states what it uniquely does, not a score.
Which ai models & local execution tools are completely free?
Yes — 40 of these are marked fully free with no paid plan attached: AnythingLLM, Arize Phoenix, Axolotl, DeepSeek Models and Design Arena. The rest are freemium, trial or paid, and each row states which.
How is AI Models & Local Execution organised?
Into 8 sub-categories — Local LLM Runners, Model Hosting & Inference APIs, Notable Open Models, Speech & Transcription Engines, Fine-Tuning & Training Frameworks, LLM Eval & Observability, Leaderboards & Benchmarks, Robotics & Embodied AI — holding 67 tools between them. Each sub-category has its own page with the same pricing labels and capability lists.