# The Local Bench

> A datasheet magazine for local AI: open-weight models, home hardware, measured benchmarks and an honest cost comparison with subscriptions. Every value carries a source and a flag: verified (checked against config.json / HF API), sourced, estimate (formula) or sample (placeholder). Data as of 2026-10-12. Licence: CC BY 4.0. Cite as "The Local Bench (thelocalbench.com), data as of 2026-10-12, CC BY 4.0".

Every HTML page has a Markdown twin at the same path with ".md" (the home page is /index.md).

## Start here

- [Methodology, labels and corrections](https://thelocalbench.com/about.md): formulas, flags, corrections policy
- [Open data](https://thelocalbench.com/data.md): downloads and licence

## Data

- [OpenAPI description](https://thelocalbench.com/api/v1/openapi.json): static read-only JSON API, CORS open
- [Models JSON](https://thelocalbench.com/api/v1/models.json)
- [Hardware JSON](https://thelocalbench.com/api/v1/hardware.json)
- [Benchmarks JSON](https://thelocalbench.com/api/v1/benchmarks.json)
- [Special-purpose models JSON](https://thelocalbench.com/api/v1/specialists.json): 102 models in 14 categories
- [Use cases JSON](https://thelocalbench.com/api/v1/use-cases.json)
- [Tools JSON](https://thelocalbench.com/api/v1/tools.json): 40 tools that work with local models; each has a machine-readable sponsor field (sponsored, disclosure)
- [New tools this month](https://thelocalbench.com/tools/new.json): JSON Feed, also [RSS](https://thelocalbench.com/tools/new.xml)
- [Full catalogue JSON](https://thelocalbench.com/data/latest/catalog.json)
- [Models CSV](https://thelocalbench.com/data/latest/models.csv)

## Explainers

- [What do context length, RoPE and YaRN mean, and how much context do you really get?](https://thelocalbench.com/learn/context-length.md): Context length is how many tokens a model can consider at once. Models encode position with RoPE, and methods such as YaRN stretch it beyond the trained length: Qwen3-30B-A3B is 32,768 tokens native and 131,072 with YaRN. Your runtime's setting, and the KV-cache memory it needs, decide what you actually get.
- [What is a KV cache, and why does it eat my VRAM?](https://thelocalbench.com/learn/kv-cache.md): The KV cache is memory a model uses to remember every token already in the conversation, and it grows with context length. In October 2026, Qwen3 32B needs 0.26 MB per token, so 32,768 tokens add 8.6 GB to its 19.8 GB of Q4_K_M weights: too much for a 24 GB card.
- [Why is memory bandwidth the speed limit for local AI?](https://thelocalbench.com/learn/memory-bandwidth.md): Each generated token requires reading every active weight from memory once, so tokens per second can't exceed bandwidth divided by gigabytes read per token. In October 2026, a used RTX 3090 (936 GB/s) running Gemma 3 27B at Q4_K_M (16.6 GB) tops out near 57 tok/s; real runs reach about 60% of that.
- [What is a mixture-of-experts model, and why does "30B-A3B" run so fast?](https://thelocalbench.com/learn/mixture-of-experts.md): A mixture-of-experts (MoE) model splits each layer into many expert sub-networks and runs only a few per token. You must store every expert, but each token reads only the active ones, so Qwen3-30B-A3B (30.5B total, 3.3B active) needs memory like a 30B model yet decodes at speeds closer to a 3B one.
- [GGUF, EXL2/EXL3, AWQ or GPTQ — which quantized file should you download?](https://thelocalbench.com/learn/quantization-formats.md): Choose by runtime, not by name. GGUF is for llama.cpp, Ollama and LM Studio, and runs on CPUs, Macs and any GPU. EXL2 and EXL3 are for ExLlama on NVIDIA GPUs and allow any average bit rate. AWQ and GPTQ are 4-bit GPU formats for vLLM-style servers. At similar bits per weight, quality differences are small.
- [What is quantization, and how much does Q4 hurt?](https://thelocalbench.com/learn/quantization.md): Quantization stores a model's weights in fewer bits, typically 4 to 8 instead of 16. In October 2026, Gemma 3 27B shrinks from 54.0 GB in BF16 to 16.6 GB at Q4_K_M, enough to fit a 24 GB card. Quality loss is small at 5 to 8 bits, acceptable at 4, noticeable at 3.
- [What are RAG and embeddings, and how does a local model "read" your documents?](https://thelocalbench.com/learn/rag-and-embeddings.md): Retrieval-augmented generation (RAG) turns your documents into embeddings (lists of numbers that capture meaning), finds the passages closest to your question, and pastes them into the model's prompt. The model never learns your files; it reads the few relevant pieces each time. Retrieval quality, not model size, usually decides the result.
- [What is speculative decoding, and is it a free speed-up?](https://thelocalbench.com/learn/speculative-decoding.md): Speculative decoding lets a small, fast draft model guess the next few tokens, then has the large model check all the guesses in one pass. Accepted guesses come out almost for free, so the output is identical to the large model's own, but typically 1.5–3× faster when the draft guesses well.
- [What are tool calling and MCP, and can a local model use them?](https://thelocalbench.com/learn/tool-calling-and-mcp.md): With tool calling, the model replies with a structured request (a function name plus JSON arguments) instead of prose. Your app runs the function and returns the result. MCP is an open protocol, built on JSON-RPC 2.0, that lets any compatible app plug into servers offering tools. Local models handle it well if they were trained for it.
- [Glossary](https://thelocalbench.com/learn.md): 29 terms

## How-to guides

- [Run your first local model with Ollama and Open WebUI](https://thelocalbench.com/how-to/first-local-model-ollama-open-webui.md): Install Ollama, pull Gemma 3, check the local API answers on port 11434, and run Open WebUI in Docker on localhost:3000 so you can chat in a browser. Use local models and disable external connections for local-only inference. (Sourced, not yet tested)
- [Build llama.cpp and run a model with GPU offload](https://thelocalbench.com/how-to/llama-cpp-gpu-offload.md): Clone and build llama.cpp with -DGGML_CUDA=ON, start llama-server with -hf and -ngl, check /health and the OpenAI-compatible chat endpoint, then tune -ngl, -c and -fa to fit your VRAM. (Partly tested)
- [Connect a local model to a coding agent (Aider + Ollama)](https://thelocalbench.com/how-to/local-model-coding-agent-aider.md): Install Aider with aider-install, point it at Ollama with OLLAMA_API_BASE, pull a coding model such as qwen3-coder:30b, start Ollama with a larger context, and run aider --model ollama_chat/<model> inside a git repository. (Sourced, not yet tested)
- [Set up private document search (RAG) with Ollama and Open WebUI](https://thelocalbench.com/how-to/private-document-search-rag.md): Pull a local embedding model and a tool-capable chat model in Ollama, raise Ollama's context length, point Open WebUI's document embedding at Ollama, create a knowledge base, upload files, and query it with (Sourced, not yet tested)
- [Run a tiny local model on your Mac CPU](https://thelocalbench.com/how-to/tiny-model-cpu.md): Run the TinyStories 15M model with a pinned llama2.c runtime on macOS. Verify the download checksum, compile with Apple clang, and generate text on the CPU. (Tested by us)
- [How-to JSON](https://thelocalbench.com/api/v1/how-to.json)

## Issue

- [Issue 001: The 24 GB question](https://thelocalbench.com/issues/001.md): October 2026 cover story and state of local AI, every figure cited

## Tools directory

- [Tools for local models](https://thelocalbench.com/tools.md): runtimes, chat apps, coding agents, frameworks, design and ops tools, with confirmed Ollama / llama.cpp / LM Studio / OpenAI-compatible / MCP support and sources. Order is editorial; no paid placement.

## Use cases

- [A private voice assistant for the home](https://thelocalbench.com/use-cases/private-voice-assistant.md): Wake word, speech recognition, a small language model and a voice, all on a box in the cupboard. Nothing you say leaves the house.
- [Offline document search with RAG](https://thelocalbench.com/use-cases/offline-document-search.md): Ask questions of your own PDFs, notes and manuals, with citations, on a machine that never phones home.
- [A local coding copilot](https://thelocalbench.com/use-cases/coding-copilot.md): Autocomplete in the editor and an agent that edits files and runs tests, with your code staying on your machine.
- [Receipt and paperwork scanning](https://thelocalbench.com/use-cases/receipt-scanning.md): Photograph receipts and letters, get searchable text, titles and tags, and keep the archive on your own disk.
- [Search your photo library by description](https://thelocalbench.com/use-cases/photo-search.md): Type "dog on the beach at sunset" and find it, across 20 years of photos, without a cloud photo service.
- [Smarter camera alerts](https://thelocalbench.com/use-cases/camera-alerts.md): Get "a courier left a parcel at the door" instead of "motion detected", with the video never leaving your network.
- [A medical literature helper (not medical advice)](https://thelocalbench.com/use-cases/medical-literature-helper.md): Search and summarise PubMed papers with a medical model, privately, as a reading aid for clinicians, researchers and curious patients.
- [A patient maths tutor](https://thelocalbench.com/use-cases/math-tutoring.md): Step-by-step help with homework and exam practice on a laptop, with no accounts, no ads and no data collection on children.
- [Answering the phone for a small business](https://thelocalbench.com/use-cases/small-business-call-answering.md): A voice agent that answers, takes messages, gives opening hours and books appointments, running on hardware in the back office.
- [Teaching a robot arm at home](https://thelocalbench.com/use-cases/home-robotics.md): Record 50 demonstrations, fine-tune an open vision-language-action model, and let a desktop arm do the task by itself.
- [A private image studio](https://thelocalbench.com/use-cases/image-studio.md): Generate and edit illustrations, product shots and concept art with node-based workflows, without a subscription or a content filter you do not control.
- [Private translation for documents and speech](https://thelocalbench.com/use-cases/private-translation.md): Translate contracts, emails and recordings between dozens of languages on your own hardware.
- [Personal and small-business finance analysis](https://thelocalbench.com/use-cases/personal-finance-analysis.md): Categorise bank exports, ask questions in plain English and forecast cash flow, with your statements never leaving your machine.

## Special-purpose model rankings

- [Math and reasoning](https://thelocalbench.com/models/c/math-reasoning.md): ranked by AIME 2024, pass@1
- [Coding](https://thelocalbench.com/models/c/coding.md): ranked by SWE-bench Verified, % resolved
- [Vision-language](https://thelocalbench.com/models/c/vision-language.md): ranked by MMMU (val)
- [OCR and documents](https://thelocalbench.com/models/c/ocr-documents.md): ranked by olmOCR-Bench, overall
- [Speech to text](https://thelocalbench.com/models/c/speech-to-text.md): ranked by Open ASR Leaderboard, average WER (English short-form)
- [Text to speech](https://thelocalbench.com/models/c/text-to-speech.md): ranked by No common public benchmark captured yet
- [Image generation](https://thelocalbench.com/models/c/image-generation.md): ranked by GenEval, overall
- [Embeddings and rerankers](https://thelocalbench.com/models/c/embeddings-rerankers.md): ranked by MMTEB (multilingual) mean task score; rerankers: MTEB-R
- [Small fast 'System 1'](https://thelocalbench.com/models/c/small-fast.md): ranked by BFCL V4, overall accuracy
- [Biomedical](https://thelocalbench.com/models/c/biomedical.md): ranked by MedQA (4-option), accuracy
- [Tool use and agents](https://thelocalbench.com/models/c/tool-use-agents.md): ranked by BFCL V4, overall accuracy
- [Multilingual and translation](https://thelocalbench.com/models/c/multilingual.md): ranked by LMArena text, non-English category (Elo-style)
- [Time series](https://thelocalbench.com/models/c/time-series.md): ranked by GIFT-Eval MASE, geometric mean relative to seasonal naive (lower is better)
- [Robotics (VLA)](https://thelocalbench.com/models/c/robotics-vla.md): ranked by LIBERO, average success rate (4 suites)

## Rankings

- [Rankings by memory class](https://thelocalbench.com/rankings.md)
- [Configurator URL parameters](https://thelocalbench.com/configurator.md)

## Optional

- [Llama 3.1 8B Instruct](https://thelocalbench.com/models/llama-3.1-8b.md)
- [Qwen3 8B](https://thelocalbench.com/models/qwen3-8b.md)
- [Qwen3 14B](https://thelocalbench.com/models/qwen3-14b.md)
- [gpt-oss-20b](https://thelocalbench.com/models/gpt-oss-20b.md)
- [Mistral Small 3.2 24B](https://thelocalbench.com/models/mistral-small-3.2-24b.md)
- [Gemma 3 27B IT](https://thelocalbench.com/models/gemma-3-27b.md)
- [Qwen3 30B-A3B](https://thelocalbench.com/models/qwen3-30b-a3b.md)
- [Qwen3 32B](https://thelocalbench.com/models/qwen3-32b.md)
- [Llama 3.3 70B Instruct](https://thelocalbench.com/models/llama-3.3-70b.md)
- [gpt-oss-120b](https://thelocalbench.com/models/gpt-oss-120b.md)
- [GeForce RTX 5060 Ti 16 GB](https://thelocalbench.com/hardware/rtx-5060-ti-16g.md)
- [GeForce RTX 4060 Ti 16 GB](https://thelocalbench.com/hardware/rtx-4060-ti-16g.md)
- [Radeon RX 7900 XTX 24 GB](https://thelocalbench.com/hardware/rx-7900-xtx-24g.md)
- [GeForce RTX 3090 24 GB (used)](https://thelocalbench.com/hardware/rtx-3090-24g-used.md)
- [CPU only · Ryzen 9 9950X + 128 GB DDR5](https://thelocalbench.com/hardware/cpu-ryzen-9-9950x-128g.md)
- [2× RTX 3090 (used) 48 GB](https://thelocalbench.com/hardware/2x-rtx-3090-48g-used.md)
- [Mac Studio M5 Max (base)](https://thelocalbench.com/hardware/mac-studio-m5-max.md)
- [Ryzen AI Max+ 395 mini PC 128 GB](https://thelocalbench.com/hardware/ryzen-ai-max-395-128g.md)
- [GeForce RTX 4090 24 GB](https://thelocalbench.com/hardware/rtx-4090-24g.md)
- [NVIDIA DGX Spark FE 128 GB](https://thelocalbench.com/hardware/dgx-spark-128g.md)
- [MacBook Pro M4 Max 40c · 128 GB](https://thelocalbench.com/hardware/macbook-pro-m4-max-128g.md)
- [Mac Studio M5 Ultra (base)](https://thelocalbench.com/hardware/mac-studio-m5-ultra.md)
- [GeForce RTX 5090 32 GB](https://thelocalbench.com/hardware/rtx-5090-32g.md)
- [Mac Studio M5 Ultra · 512 GB](https://thelocalbench.com/hardware/mac-studio-m5-ultra-512g.md)
- [Value 16 GB build](https://thelocalbench.com/builds/value-16gb.md)
- [Used-3090 24 GB build](https://thelocalbench.com/builds/used-3090-24gb.md)
- [Mac Studio M5 Max build](https://thelocalbench.com/builds/mac-studio-unified.md)
- [Strix Halo SFF 128 GB build](https://thelocalbench.com/builds/strix-halo-128gb.md)
- [Dual-GPU 48 GB build](https://thelocalbench.com/builds/dual-gpu-48gb.md)
- [Silent home server build](https://thelocalbench.com/builds/silent-home-server.md)
- [DeepSeek R1-0528 Qwen3 8B](https://thelocalbench.com/models/s/deepseek-r1-0528-qwen3-8b.md)
- [OpenReasoning Nemotron 32B](https://thelocalbench.com/models/s/openreasoning-nemotron-32b.md)
- [QwQ 32B](https://thelocalbench.com/models/s/qwq-32b.md)
- [Phi-4 reasoning](https://thelocalbench.com/models/s/phi-4-reasoning.md)
- [OpenMath Nemotron 7B](https://thelocalbench.com/models/s/openmath-nemotron-7b.md)
- [Phi-4 mini reasoning](https://thelocalbench.com/models/s/phi-4-mini-reasoning.md)
- [DeepSeek R1 Distill Qwen 7B](https://thelocalbench.com/models/s/deepseek-r1-distill-qwen-7b.md)
- [Devstral Small 1.1](https://thelocalbench.com/models/s/devstral-small-2507.md)
- [OpenHands LM 32B](https://thelocalbench.com/models/s/openhands-lm-32b-v0-1.md)
- [Qwen3 Coder 30B-A3B](https://thelocalbench.com/models/s/qwen3-coder-30b-a3b-instruct.md)
- [Qwen2.5 Coder 32B](https://thelocalbench.com/models/s/qwen2-5-coder-32b-instruct.md)
- [Qwen2.5 Coder 7B](https://thelocalbench.com/models/s/qwen2-5-coder-7b-instruct.md)
- [DeepSeek Coder V2 Lite](https://thelocalbench.com/models/s/deepseek-coder-v2-lite-instruct.md)
- [Qwen2.5 VL 7B](https://thelocalbench.com/models/s/qwen2-5-vl-7b-instruct.md)
- [Kimi-VL A3B](https://thelocalbench.com/models/s/kimi-vl-a3b-instruct.md)
- [SmolVLM2 2.2B](https://thelocalbench.com/models/s/smolvlm2-2-2b-instruct.md)
- [Qwen3 VL 8B](https://thelocalbench.com/models/s/qwen3-vl-8b-instruct.md)
- [Qwen3 VL 30B-A3B](https://thelocalbench.com/models/s/qwen3-vl-30b-a3b-instruct.md)
- [InternVL3 8B](https://thelocalbench.com/models/s/internvl3-8b.md)
- [Gemma 3 12B](https://thelocalbench.com/models/s/gemma-3-12b-it.md)
- [olmOCR 2 7B](https://thelocalbench.com/models/s/olmocr-2-7b-1025.md)
- [PaddleOCR-VL](https://thelocalbench.com/models/s/paddleocr-vl.md)
- [DeepSeek-OCR](https://thelocalbench.com/models/s/deepseek-ocr.md)
- [Nanonets OCR2 3B](https://thelocalbench.com/models/s/nanonets-ocr2-3b.md)
- [dots.ocr](https://thelocalbench.com/models/s/dots-ocr.md)
- [Granite Docling 258M](https://thelocalbench.com/models/s/granite-docling-258m.md)
- [SmolDocling 256M](https://thelocalbench.com/models/s/smoldocling-256m-preview.md)
- [Canary-Qwen 2.5B](https://thelocalbench.com/models/s/canary-qwen-2-5b.md)
- [Parakeet TDT 0.6B v2](https://thelocalbench.com/models/s/parakeet-tdt-0-6b-v2.md)
- [Parakeet TDT 0.6B v3](https://thelocalbench.com/models/s/parakeet-tdt-0-6b-v3.md)
- [Distil-Whisper large-v3.5](https://thelocalbench.com/models/s/distil-large-v3-5.md)
- [Voxtral Mini 3B](https://thelocalbench.com/models/s/voxtral-mini-3b-2507.md)
- [Whisper large-v3](https://thelocalbench.com/models/s/whisper-large-v3.md)
- [Whisper large-v3 turbo](https://thelocalbench.com/models/s/whisper-large-v3-turbo.md)
- [Canary 1B Flash](https://thelocalbench.com/models/s/canary-1b-flash.md)
- [Kokoro 82M](https://thelocalbench.com/models/s/kokoro-82m.md)
- [Chatterbox](https://thelocalbench.com/models/s/chatterbox.md)
- [CSM 1B](https://thelocalbench.com/models/s/csm-1b.md)
- [Dia 1.6B](https://thelocalbench.com/models/s/dia-1-6b.md)
- [Orpheus 3B](https://thelocalbench.com/models/s/orpheus-3b-0-1-ft.md)
- [VibeVoice 1.5B](https://thelocalbench.com/models/s/vibevoice-1-5b.md)
- [OpenAudio S1 mini](https://thelocalbench.com/models/s/openaudio-s1-mini.md)
- [HiDream-I1 Full](https://thelocalbench.com/models/s/hidream-i1-full.md)
- [FLUX.1 [dev]](https://thelocalbench.com/models/s/flux-1-dev.md)
- [Stable Diffusion XL](https://thelocalbench.com/models/s/stable-diffusion-xl-base-1-0.md)
- [FLUX.1 [schnell]](https://thelocalbench.com/models/s/flux-1-schnell.md)
- [FLUX.1 Kontext [dev]](https://thelocalbench.com/models/s/flux-1-kontext-dev.md)
- [Qwen-Image](https://thelocalbench.com/models/s/qwen-image.md)
- [Stable Diffusion 3.5 Medium](https://thelocalbench.com/models/s/stable-diffusion-3-5-medium.md)
- [Qwen3 Embedding 8B](https://thelocalbench.com/models/s/qwen3-embedding-8b.md)
- [Qwen3 Embedding 0.6B](https://thelocalbench.com/models/s/qwen3-embedding-0-6b.md)
- [multilingual-e5-large-instruct](https://thelocalbench.com/models/s/multilingual-e5-large-instruct.md)
- [BGE-M3](https://thelocalbench.com/models/s/bge-m3.md)
- [EmbeddingGemma 300M](https://thelocalbench.com/models/s/embeddinggemma-300m.md)
- [Nomic Embed Text v1.5](https://thelocalbench.com/models/s/nomic-embed-text-v1-5.md)
- [jina-embeddings-v3](https://thelocalbench.com/models/s/jina-embeddings-v3.md)
- [Qwen3 Reranker 0.6B](https://thelocalbench.com/models/s/qwen3-reranker-0-6b.md)
- [BGE Reranker v2 M3](https://thelocalbench.com/models/s/bge-reranker-v2-m3.md)
- [mxbai-rerank-large v2](https://thelocalbench.com/models/s/mxbai-rerank-large-v2.md)
- [Qwen3 0.6B](https://thelocalbench.com/models/s/qwen3-0-6b.md)
- [Llama 3.2 1B](https://thelocalbench.com/models/s/llama-3-2-1b-instruct.md)
- [Gemma 3 1B](https://thelocalbench.com/models/s/gemma-3-1b-it.md)
- [Gemma 3 270M](https://thelocalbench.com/models/s/gemma-3-270m.md)
- [Qwen3 1.7B](https://thelocalbench.com/models/s/qwen3-1-7b.md)
- [SmolLM3 3B](https://thelocalbench.com/models/s/smollm3-3b.md)
- [LFM2 1.2B](https://thelocalbench.com/models/s/lfm2-1-2b.md)
- [Phi-4 mini](https://thelocalbench.com/models/s/phi-4-mini-instruct.md)
- [Granite 4.0 H Micro](https://thelocalbench.com/models/s/granite-4-0-h-micro.md)
- [MedGemma 27B](https://thelocalbench.com/models/s/medgemma-27b-it.md)
- [MedGemma 1.5 4B](https://thelocalbench.com/models/s/medgemma-1-5-4b-it.md)
- [MedGemma 4B](https://thelocalbench.com/models/s/medgemma-4b-it.md)
- [MedGemma 27B text](https://thelocalbench.com/models/s/medgemma-27b-text-it.md)
- [TxGemma 9B chat](https://thelocalbench.com/models/s/txgemma-9b-chat.md)
- [II-Medical 8B](https://thelocalbench.com/models/s/ii-medical-8b.md)
- [Med42 8B](https://thelocalbench.com/models/s/llama3-med42-8b.md)
- [OpenBioLLM 8B](https://thelocalbench.com/models/s/llama3-openbiollm-8b.md)
- [xLAM-2 8B](https://thelocalbench.com/models/s/llama-xlam-2-8b-fc-r.md)
- [ToolACE-2 8B](https://thelocalbench.com/models/s/toolace-2-8b.md)
- [Qwen3 4B Instruct 2507](https://thelocalbench.com/models/s/qwen3-4b-instruct-2507.md)
- [GLM-4.5 Air](https://thelocalbench.com/models/s/glm-4-5-air.md)
- [watt-tool 8B](https://thelocalbench.com/models/s/watt-tool-8b.md)
- [Hermes 3 8B](https://thelocalbench.com/models/s/hermes-3-llama-3-1-8b.md)
- [Arch-Function 3B](https://thelocalbench.com/models/s/arch-function-3b.md)
- [Gemma 3 4B](https://thelocalbench.com/models/s/gemma-3-4b-it.md)
- [Aya Expanse 32B](https://thelocalbench.com/models/s/aya-expanse-32b.md)
- [Aya Expanse 8B](https://thelocalbench.com/models/s/aya-expanse-8b.md)
- [Hunyuan-MT 7B](https://thelocalbench.com/models/s/hunyuan-mt-7b.md)
- [Seed-X PPO 7B](https://thelocalbench.com/models/s/seed-x-ppo-7b.md)
- [EuroLLM 9B](https://thelocalbench.com/models/s/eurollm-9b-instruct.md)
- [Tower+ 9B](https://thelocalbench.com/models/s/tower-plus-9b.md)
- [NLLB-200 600M](https://thelocalbench.com/models/s/nllb-200-distilled-600m.md)
- [Chronos-2](https://thelocalbench.com/models/s/chronos-2.md)
- [TimesFM 2.5](https://thelocalbench.com/models/s/timesfm-2-5-200m-pytorch.md)
- [Moirai 2.0 small](https://thelocalbench.com/models/s/moirai-2-0-r-small.md)
- [TimesFM 2.0 500M](https://thelocalbench.com/models/s/timesfm-2-0-500m-pytorch.md)
- [Chronos-Bolt base](https://thelocalbench.com/models/s/chronos-bolt-base.md)
- [Chronos-Bolt small](https://thelocalbench.com/models/s/chronos-bolt-small.md)
- [Granite TTM r2](https://thelocalbench.com/models/s/granite-timeseries-ttm-r2.md)
- [SmolVLA](https://thelocalbench.com/models/s/smolvla-base.md)
- [π0 (LeRobot port)](https://thelocalbench.com/models/s/pi0.md)
- [OpenVLA 7B](https://thelocalbench.com/models/s/openvla-7b.md)
- [GR00T N1.5 3B](https://thelocalbench.com/models/s/gr00t-n1-5-3b.md)