llms.txt
Compare with other sites' llms.txt files →# llmrun
> llmrun answers "can I run this LLM locally?". For a model and a GPU, Apple Silicon Mac or other device it estimates the VRAM needed per quantization, decode speed (tok/s, with MoE models computed from active parameters) and prefill time (TTFT). It also collects third-party benchmark results with attribution and publishes the llmrun Score, a ranking built from them.
Speed and fit figures are estimates from a bandwidth and compute model, not live measurements. The how-it-works page lists what they leave out. Numbers change as models, hardware and benchmarks are added, so read the page for current values and do not cache them. When citing, link the page you used.
## Tools
- [Home](https://llmrun.dev): Pick a GPU or device and see which models fit and how fast they run
- [Cloud GPU rental](https://llmrun.dev/cloud): Rental prices for GPUs when your hardware is too small
## Hardware
- [GPUs](https://llmrun.dev/gpu): All GPUs with VRAM, bandwidth and compatible models
- [Devices](https://llmrun.dev/device): Laptops, desktops, Apple Silicon Macs and AI boxes
- [GPUs with 8 GB VRAM](https://llmrun.dev/gpu/tier/8gb-vram): Which models fit in 8 GB VRAM
- [GPUs with 12 GB VRAM](https://llmrun.dev/gpu/tier/12gb-vram): Which models fit in 12 GB VRAM
- [GPUs with 16 GB VRAM](https://llmrun.dev/gpu/tier/16gb-vram): Which models fit in 16 GB VRAM
- [GPUs with 24 GB VRAM](https://llmrun.dev/gpu/tier/24gb-vram): Which models fit in 24 GB VRAM
- [GPUs with 48 GB VRAM](https://llmrun.dev/gpu/tier/48gb-vram): Which models fit in 48 GB VRAM
- [NVIDIA GeForce RTX 4090](https://llmrun.dev/gpu/rtx-4090): Models that fit, quantization options, estimated speed
- [NVIDIA GeForce RTX 5090](https://llmrun.dev/gpu/rtx-5090): Models that fit, quantization options, estimated speed
- [NVIDIA GeForce RTX 5080](https://llmrun.dev/gpu/rtx-5080): Models that fit, quantization options, estimated speed
- [NVIDIA GeForce RTX 5070 Ti](https://llmrun.dev/gpu/rtx-5070-ti): Models that fit, quantization options, estimated speed
- [NVIDIA GeForce RTX 5070](https://llmrun.dev/gpu/rtx-5070): Models that fit, quantization options, estimated speed
- [NVIDIA GeForce RTX 4080 SUPER](https://llmrun.dev/gpu/rtx-4080-super): Models that fit, quantization options, estimated speed
- [NVIDIA GeForce RTX 3090](https://llmrun.dev/gpu/rtx-3090): Models that fit, quantization options, estimated speed
- [NVIDIA GeForce RTX 4080](https://llmrun.dev/gpu/rtx-4080): Models that fit, quantization options, estimated speed
- [NVIDIA GeForce RTX 4070 Ti SUPER](https://llmrun.dev/gpu/rtx-4070-ti-super): Models that fit, quantization options, estimated speed
- [AMD Radeon RX 9070 XT](https://llmrun.dev/gpu/rx-9070-xt): Models that fit, quantization options, estimated speed
- [NVIDIA GeForce RTX 4070 SUPER](https://llmrun.dev/gpu/rtx-4070-super): Models that fit, quantization options, estimated speed
- [NVIDIA RTX PRO 6000 Blackwell Workstation Edition](https://llmrun.dev/gpu/rtx-pro-6000-blackwell): Models that fit, quantization options, estimated speed
- [NVIDIA GeForce RTX 4070 Ti](https://llmrun.dev/gpu/rtx-4070-ti): Models that fit, quantization options, estimated speed
- [NVIDIA GeForce RTX 5060 Ti 16GB](https://llmrun.dev/gpu/rtx-5060-ti-16gb): Models that fit, quantization options, estimated speed
- [AMD Radeon RX 9070](https://llmrun.dev/gpu/rx-9070): Models that fit, quantization options, estimated speed
- [AMD Radeon RX 9070 GRE](https://llmrun.dev/gpu/rx-9070-gre): Models that fit, quantization options, estimated speed
- [NVIDIA GeForce RTX 4060](https://llmrun.dev/gpu/rtx-4060): Models that fit, quantization options, estimated speed
- [NVIDIA GeForce RTX 4070](https://llmrun.dev/gpu/rtx-4070): Models that fit, quantization options, estimated speed
- [AMD Radeon RX 7900 XTX](https://llmrun.dev/gpu/rx-7900-xtx): Models that fit, quantization options, estimated speed
- [NVIDIA H200 SXM](https://llmrun.dev/gpu/h200-sxm): Models that fit, quantization options, estimated speed
- [Mac Studio M4 Max (128 GB)](https://llmrun.dev/device/mac-studio-m4-max-128gb): Models that fit in its memory, estimated speed
- [MacBook Pro 16" M4 Max (64 GB)](https://llmrun.dev/device/macbook-pro-16-m4-max-64gb): Models that fit in its memory, estimated speed
- [Mac Studio M4 Max (64 GB)](https://llmrun.dev/device/mac-studio-m4-max-64gb): Models that fit in its memory, estimated speed
- [Mac Studio (M3 Ultra, 512GB)](https://llmrun.dev/device/mac-studio-m3-ultra-512gb): Models that fit in its memory, estimated speed
- [Mac Mini M4 Pro (24 GB)](https://llmrun.dev/device/mac-mini-m4-pro-24gb): Models that fit in its memory, estimated speed
- [MacBook Pro 16" M4 Max (48 GB)](https://llmrun.dev/device/macbook-pro-16-m4-max-48gb): Models that fit in its memory, estimated speed
- [MacBook Pro 16" M5 Max (128 GB)](https://llmrun.dev/device/macbook-pro-m5-max): Models that fit in its memory, estimated speed
- [Mac Mini M4 (16 GB)](https://llmrun.dev/device/mac-mini-m4-16gb): Models that fit in its memory, estimated speed
- [MacBook Pro 14" M4 Pro (24 GB)](https://llmrun.dev/device/macbook-pro-14-m4-pro-24gb): Models that fit in its memory, estimated speed
- [Mac Studio (M3 Ultra, 256GB)](https://llmrun.dev/device/mac-studio-m3-ultra-256gb): Models that fit in its memory, estimated speed
## Models
- [Models](https://llmrun.dev/model): Browse and filter models by size, capability and VRAM
- [Best local LLMs for coding](https://llmrun.dev/use/coding): Models suited to coding and what they need to run
- [Best local LLMs for reasoning](https://llmrun.dev/use/reasoning): Models suited to reasoning and what they need to run
- [Best local LLMs for vision](https://llmrun.dev/use/vision): Models suited to vision and what they need to run
- [Best local LLMs for tool use](https://llmrun.dev/use/tool-use): Models suited to tool use and what they need to run
- [Best local LLMs for long context](https://llmrun.dev/use/long-context): Models suited to long context and what they need to run
- [Best local LLMs for small models](https://llmrun.dev/use/small-models): Models suited to small models and what they need to run
- [Qwen3.8 27B](https://llmrun.dev/model/qwen-qwen3-8-27b): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen3.6 35B A3B](https://llmrun.dev/model/qwen-qwen3-6-35b-a3b): VRAM per quantization, estimated tok/s, GPUs that run it
- [Gemma 4 26B A4B IT](https://llmrun.dev/model/google-gemma-4-26b-a4b-it): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen3.5 9B](https://llmrun.dev/model/qwen-qwen3-5-9b): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen3 8B](https://llmrun.dev/model/qwen-qwen3-8b): VRAM per quantization, estimated tok/s, GPUs that run it
- [Gemma 4 E4B IT](https://llmrun.dev/model/google-gemma-4-e4b-it): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen2.5 VL 7B Instruct](https://llmrun.dev/model/qwen-qwen2-5-vl-7b-instruct): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen2.5 7B Instruct](https://llmrun.dev/model/qwen-qwen2-5-7b-instruct): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen3.6 27B](https://llmrun.dev/model/qwen-qwen3-6-27b): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen3 14B](https://llmrun.dev/model/qwen-qwen3-14b): VRAM per quantization, estimated tok/s, GPUs that run it
- [Gemma 4 12B IT](https://llmrun.dev/model/google-gemma-4-12b-it): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen3.5 4B](https://llmrun.dev/model/qwen-qwen3-5-4b): VRAM per quantization, estimated tok/s, GPUs that run it
- [Gemma 4 31B IT](https://llmrun.dev/model/google-gemma-4-31b-it): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen2.5 32B Instruct](https://llmrun.dev/model/qwen-qwen2-5-32b-instruct): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen2.5 Coder 7B Instruct](https://llmrun.dev/model/qwen-qwen2-5-coder-7b-instruct): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen3 4B Instruct 2507](https://llmrun.dev/model/qwen-qwen3-4b-instruct-2507): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen3.8 Flash Next](https://llmrun.dev/model/qwen-qwen3-8-flash-next): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen3 VL 8B Instruct](https://llmrun.dev/model/qwen-qwen3-vl-8b-instruct): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen3 Coder 30B A3B Instruct](https://llmrun.dev/model/qwen-qwen3-coder-30b-a3b-instruct): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen3 4B](https://llmrun.dev/model/qwen-qwen3-4b): VRAM per quantization, estimated tok/s, GPUs that run it
- [Llama 3.2 1B Instruct](https://llmrun.dev/model/meta-llama-llama-3-2-1b-instruct): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen2.5 14B Instruct](https://llmrun.dev/model/qwen-qwen2-5-14b-instruct): VRAM per quantization, estimated tok/s, GPUs that run it
- [NVIDIA Nemotron 3 Nano 30B A3B BF16](https://llmrun.dev/model/nvidia-nvidia-nemotron-3-nano-30b-a3b-bf16): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen2.5 3B Instruct](https://llmrun.dev/model/qwen-qwen2-5-3b-instruct): VRAM per quantization, estimated tok/s, GPUs that run it
- [Gemma 4 E2B IT](https://llmrun.dev/model/google-gemma-4-e2b-it): VRAM per quantization, estimated tok/s, GPUs that run it
- [DeepSeek V4.1 Flash](https://llmrun.dev/model/deepseek-ai-deepseek-v4-1-flash): VRAM per quantization, estimated tok/s, GPUs that run it
- [NVIDIA Nemotron 3.5 Lightning 30B A3B BF16](https://llmrun.dev/model/nvidia-nvidia-nemotron-3-5-lightning-30b-a3b-bf16): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen3 VL 4B Instruct](https://llmrun.dev/model/qwen-qwen3-vl-4b-instruct): VRAM per quantization, estimated tok/s, GPUs that run it
- [Gemma 4 12B IT Qat Q4 0 Unquantized](https://llmrun.dev/model/google-gemma-4-12b-it-qat-q4-0-unquantized): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen2.5 Coder 32B Instruct](https://llmrun.dev/model/qwen-qwen2-5-coder-32b-instruct): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen3 30B A3B Instruct 2507](https://llmrun.dev/model/qwen-qwen3-30b-a3b-instruct-2507): VRAM per quantization, estimated tok/s, GPUs that run it
- [DeepSeek V4 Flash](https://llmrun.dev/model/deepseek-ai-deepseek-v4-flash): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen2 VL 7B Instruct](https://llmrun.dev/model/qwen-qwen2-vl-7b-instruct): VRAM per quantization, estimated tok/s, GPUs that run it
- [Muse Glimmer 30B](https://llmrun.dev/model/meta-models-muse-glimmer-30b): VRAM per quantization, estimated tok/s, GPUs that run it
- [Llama 3.1 8B Instruct](https://llmrun.dev/model/meta-llama-llama-3-1-8b-instruct): VRAM per quantization, estimated tok/s, GPUs that run it
- [Hy MT2 7B](https://llmrun.dev/model/tencent-hy-mt2-7b): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen2.5 VL 3B Instruct](https://llmrun.dev/model/qwen-qwen2-5-vl-3b-instruct): VRAM per quantization, estimated tok/s, GPUs that run it
- [GPT OSS 20B](https://llmrun.dev/model/openai-gpt-oss-20b): VRAM per quantization, estimated tok/s, GPUs that run it
- [MiMo V2.6 Distill Qwen 9B](https://llmrun.dev/model/xiaomimimo-mimo-v2-6-distill-qwen-9b): VRAM per quantization, estimated tok/s, GPUs that run it
- [Qwen3 32B](https://llmrun.dev/model/qwen-qwen3-32b): VRAM per quantization, estimated tok/s, GPUs that run it
## Benchmarks
- [Benchmarks](https://llmrun.dev/benchmark): Third-party benchmark results gathered in one place, with sources
- [llmrun Score](https://llmrun.dev/benchmark/llmrun-score): llmrun's combined ranking built from external benchmarks
## Methodology
- [How it works](https://llmrun.dev/how-it-works): How VRAM, decode speed, prefill/TTFT and Fit grades are estimated
- [llmrun Score methodology](https://llmrun.dev/benchmark/llmrun-score/methodology): How the score is computed from benchmark data
- [About the benchmarks](https://llmrun.dev/benchmark/about): Where benchmark data comes from and how it is credited
## Optional
- [Qwen 3.8 family](https://llmrun.dev/family/qwen-3-8): Sizes and variants in this model family
- [Qwen 3 family](https://llmrun.dev/family/qwen-3): Sizes and variants in this model family
- [Qwen 2.5 family](https://llmrun.dev/family/qwen-2-5): Sizes and variants in this model family
- [Gemma 4 family](https://llmrun.dev/family/gemma-4): Sizes and variants in this model family
- [Qwen 3.6 family](https://llmrun.dev/family/qwen-3-6): Sizes and variants in this model family
- [Qwen 3.5 family](https://llmrun.dev/family/qwen-3-5): Sizes and variants in this model family
- [Llama 3 family](https://llmrun.dev/family/llama-3): Sizes and variants in this model family
- [Nemotron family](https://llmrun.dev/family/nemotron): Sizes and variants in this model family
- [DeepSeek V4 family](https://llmrun.dev/family/deepseek-v4): Sizes and variants in this model family
- [Gemma 3 family](https://llmrun.dev/family/gemma-3): Sizes and variants in this model family
- [LFM2.5 family](https://llmrun.dev/family/lfm2-5): Sizes and variants in this model family
- [Mistral family](https://llmrun.dev/family/mistral): Sizes and variants in this model family
- [DeepSeek R1 family](https://llmrun.dev/family/deepseek-r1): Sizes and variants in this model family
- [GLM 5 family](https://llmrun.dev/family/glm-5): Sizes and variants in this model family
- [GPT-OSS family](https://llmrun.dev/family/gpt-oss): Sizes and variants in this model family
- [Qwen 2 family](https://llmrun.dev/family/qwen-2): Sizes and variants in this model family
- [GLM 4 family](https://llmrun.dev/family/glm-4): Sizes and variants in this model family
- [Hunyuan MT2 family](https://llmrun.dev/family/hunyuan-mt2): Sizes and variants in this model family
- [MiMo family](https://llmrun.dev/family/mimo): Sizes and variants in this model family
- [Ling family](https://llmrun.dev/family/ling): Sizes and variants in this model family
- [MiniCPM family](https://llmrun.dev/family/minicpm): Sizes and variants in this model family
- [Inkling family](https://llmrun.dev/family/inkling): Sizes and variants in this model family
- [Ornith family](https://llmrun.dev/family/ornith): Sizes and variants in this model family
- [Kimi K2 family](https://llmrun.dev/family/kimi-k2): Sizes and variants in this model family
- [Gemma 2 family](https://llmrun.dev/family/gemma-2): Sizes and variants in this model family
- [Granite family](https://llmrun.dev/family/granite): Sizes and variants in this model family
- [Mellum family](https://llmrun.dev/family/mellum): Sizes and variants in this model family
- [Phi 4 family](https://llmrun.dev/family/phi-4): Sizes and variants in this model family
- [SmolLM family](https://llmrun.dev/family/smollm): Sizes and variants in this model family
- [EXAONE family](https://llmrun.dev/family/exaone): Sizes and variants in this model family
- [GPQA Diamond](https://llmrun.dev/benchmark/gpqa-diamond): Leaderboard for this benchmark
- [ARC-AGI](https://llmrun.dev/benchmark/arc-agi): Leaderboard for this benchmark
- [ARC-AGI-2](https://llmrun.dev/benchmark/arc-agi-2): Leaderboard for this benchmark
- [LiveBench Reasoning](https://llmrun.dev/benchmark/livebench-reasoning): Leaderboard for this benchmark
- [LiveBench 2026 Reasoning](https://llmrun.dev/benchmark/livebench2-reasoning): Leaderboard for this benchmark
- [BIG-Bench Hard](https://llmrun.dev/benchmark/bbh): Leaderboard for this benchmark
- [SimpleBench](https://llmrun.dev/benchmark/simplebench): Leaderboard for this benchmark
- [CritPt](https://llmrun.dev/benchmark/critpt): Leaderboard for this benchmark
- [WinoGrande](https://llmrun.dev/benchmark/winogrande): Leaderboard for this benchmark
- [PIQA](https://llmrun.dev/benchmark/piqa): Leaderboard for this benchmark
- [APEX-Agents](https://llmrun.dev/benchmark/apex-agents): Leaderboard for this benchmark
- [Chess Puzzles](https://llmrun.dev/benchmark/chess-puzzles): Leaderboard for this benchmark
- [Mystery Game Puzzles](https://llmrun.dev/benchmark/mystery-game-puzzles): Leaderboard for this benchmark
- [DTBench](https://llmrun.dev/benchmark/dtbench): Leaderboard for this benchmark
- [LMCA](https://llmrun.dev/benchmark/lmca): Leaderboard for this benchmark
- [BALROG](https://llmrun.dev/benchmark/balrog): Leaderboard for this benchmark
- [Fiction.LiveBench](https://llmrun.dev/benchmark/fiction-livebench): Leaderboard for this benchmark
- [CL-bench](https://llmrun.dev/benchmark/cl-bench): Leaderboard for this benchmark
- [CL-bench Life](https://llmrun.dev/benchmark/cl-bench-life): Leaderboard for this benchmark
- [LiveBench 2026 Data Analysis](https://llmrun.dev/benchmark/livebench2-data-analysis): Leaderboard for this benchmark
- [LiveBench 2026 Browser](https://llmrun.dev/benchmark/livebench2-browser): Leaderboard for this benchmark
- [ForecastBench](https://llmrun.dev/benchmark/forecastbench): Leaderboard for this benchmark
- [Vending-Bench 2](https://llmrun.dev/benchmark/vending-bench-2): Leaderboard for this benchmark
- [Furniture Assembly](https://llmrun.dev/benchmark/furniture-assembly): Leaderboard for this benchmark
- [EBR-bench](https://llmrun.dev/benchmark/ebr-bench): Leaderboard for this benchmark
- [GDP.pdf](https://llmrun.dev/benchmark/gdp-pdf): Leaderboard for this benchmark
- [LiveBench Coding](https://llmrun.dev/benchmark/livebench-coding): Leaderboard for this benchmark
- [LiveBench 2026 Coding](https://llmrun.dev/benchmark/livebench2-coding): Leaderboard for this benchmark
- [LMArena WebDev](https://llmrun.dev/benchmark/lmarena-webdev): Leaderboard for this benchmark
- [SWE-bench Verified](https://llmrun.dev/benchmark/swe-bench-verified): Leaderboard for this benchmark
- [SWE-bench Bash Only](https://llmrun.dev/benchmark/swe-bench-bash): Leaderboard for this benchmark
- [Terminal-Bench](https://llmrun.dev/benchmark/terminalbench): Leaderboard for this benchmark
- [Aider Polyglot](https://llmrun.dev/benchmark/aider-polyglot): Leaderboard for this benchmark
- [SWE-bench Multilingual](https://llmrun.dev/benchmark/swe-bench-multilingual): Leaderboard for this benchmark
- [SWE-bench Lite](https://llmrun.dev/benchmark/swe-bench-lite): Leaderboard for this benchmark
- [SciCode](https://llmrun.dev/benchmark/scicode): Leaderboard for this benchmark
- [WeirdML](https://llmrun.dev/benchmark/weirdml): Leaderboard for this benchmark
- [FrontierCode](https://llmrun.dev/benchmark/frontiercode): Leaderboard for this benchmark
- [DeepSWE](https://llmrun.dev/benchmark/deepswe): Leaderboard for this benchmark
- [GSO-Bench](https://llmrun.dev/benchmark/gso-bench): Leaderboard for this benchmark
- [Surface Evolver Bench](https://llmrun.dev/benchmark/surface-evolver-bench): Leaderboard for this benchmark
- [FrontierSWE](https://llmrun.dev/benchmark/frontierswe): Leaderboard for this benchmark
- [LiveBench 2026 Long-Horizon](https://llmrun.dev/benchmark/livebench2-long-horizon): Leaderboard for this benchmark
- [CursorBench](https://llmrun.dev/benchmark/cursorbench): Leaderboard for this benchmark
- [ALE-Bench](https://llmrun.dev/benchmark/ale-bench): Leaderboard for this benchmark
- [ExploitBench](https://llmrun.dev/benchmark/exploitbench): Leaderboard for this benchmark
- [METR Time Horizons](https://llmrun.dev/benchmark/metr-time-horizons): Leaderboard for this benchmark
- [LiveBench 2026 Design](https://llmrun.dev/benchmark/livebench2-design): Leaderboard for this benchmark
- [WeirdML v3](https://llmrun.dev/benchmark/weirdml-v3): Leaderboard for this benchmark
- [GBA Eval](https://llmrun.dev/benchmark/gbaeval): Leaderboard for this benchmark
- [AIME 2024/2025](https://llmrun.dev/benchmark/aime): Leaderboard for this benchmark
- [MATH Level 5](https://llmrun.dev/benchmark/math-level-5): Leaderboard for this benchmark
- [LiveBench Math](https://llmrun.dev/benchmark/livebench-math): Leaderboard for this benchmark
- [FrontierMath](https://llmrun.dev/benchmark/fronti