llamaperf

Local model performance on your hardware

Find which open-weight LLMs fit in your GPU or Mac and compare the speeds people report on setups like yours.

Join the community

Follow along:XLinkedIn

What runs on your hardware?

Pick your GPU or Mac, then read the speeds people reported on it, or estimate which models fit and how fast they run.

Free to use. No account needed. Memory estimates and community measurements are labelled separately.

Looking for a particular model?

Search for a model to see its reported speeds across GPUs and Macs.

Local model performance reports from the community

These are individual setups, not a controlled benchmark. Compare GPU count, quantization, context and offloading before comparing speeds. How to read a report →

Model: Hy4-Flash
Compare setup details

Exact recorded values. Context may be a configured limit; matching filters does not establish identical prompts, offloading or concurrency.

Tone: positiveApple hardware
generation:
36.4 tokens/s
prompt processing (prefill):
968.0 tokens/s
quant:
4-bit (MLX)

User reports Hy4-Flash at 36.4 t/s decode and 968 t/s prefill on a MacBook Pro M5 Max with 128 GB, using their own Kapitsa engine. Setup is Kapitsa with a 4-bit MLX quant (8-bit trunk, 4-bit routed experts, group size 64) and the fast profile; the 134 GB pack keeps every expert in RAM with an 88 GB peak, Engram tables memory-mapped on the CPU. Decode falls from 36.4 t/s at short context to 34.6 t/s at 32K; prefill is 657 t/s at 4K on the default profile and 968-971 t/s with the fast profile. MLX 3.1-bit is compared at 20.8 t/s decode and 228 t/s prefill at 4K. A 48 GB M5 Pro streaming experts from SSD decodes at about 3.5 t/s.

Nov 20, 2026
Tone: positiveApple hardware
generation:
36.4 tokens/s
quant:
4-bit (MLX)

User reports Hy4-Flash at 36.4 t/s decode on a MacBook Pro M5 Max (40-core GPU, 128 GB), with prefill of 657 tok/s at 4K context in the default profile and 968 tok/s with --profile fast. Setup is the Kapitsa engine with a 4-bit MLX quant (8-bit trunk, 4-bit group-64 routed experts, plus Engram tables), 134 GB on disk and 88 GB peak RAM; the Engram tables are memory-mapped and read on the CPU. Decode falls from 36.4 t/s at short context to 34.6 t/s at 32K, while MLX (mlx-lm 0.32.2, 3.1-bit) drops from 20.8 to 4.4 t/s over the same contexts. Warm turns in a cached conversation start in 0.15-0.21 s at 1K to 32K. A 48 GB M5 Pro streams experts from SSD and decodes at about 3.5 tok/s.

Nov 20, 2026

Hy4-Flash

NVIDIA RTX 3060 12GB · RustingStrata · 20,000 ctx

Tone: mixedNVIDIA hardware
generation:
3.3 tokens/s
prompt processing (prefill):
11.6 tokens/s
quant:
IQ2_S/IQ2_XXS (GGUF)
coding

User reports Hy4-flash at 3.3 t/s generation and 11.6 t/s prompt reading on an RTX 3060 12GB with 32 GB of RAM, using a 20,000-token prompt and 64 tokens of output. Setup is RustingStrata with an IQ2_S/IQ2_XXS GGUF of about 85 GB, streamed from an NVMe drive with experts split between CPU and GPU. The Vulkan backend on the same card gave 1.9 t/s generation and 10.6 t/s prompt reading. After a short prompt the model writes 4.3 to 5.1 t/s on CUDA. The user notes one token needs about 2.6 GB of expert weights and calls it not suitable for a coding agent, though the full model runs on a 12 GB card. llama.cpp support was still a draft pull request, so no comparison was made.

Nov 19, 2026

Hy4-Flash

Unknown GPU

Tone: mixed
generation:
10.0 tokens/s
quant:
4-bit
long-contextsummarization

User reports Hy4-Flash at 10 t/s on a machine with 26GB VRAM and 128GB DDR4, compared with 15 t/s for QFN. Hy4-Flash ran at 4-bit and QFN at 5-bit; no engine or KV cache setting is stated. The user says Hy4-Flash is substantially more accurate at citing law from PDF documents and calls it the best model for local legal workflows, but describes the speed as poor.

Nov 19, 2026

Hy4-Flash

2× NVIDIA RTX 3090 · llama.cpp · 131,072 ctx

Tone: positiveNVIDIA hardware
generation:
20-24 tokens/s
prompt processing (prefill):
60-65 tokens/s
quant:
Q3_K (GGUF)
kv:
q8_0

User reports running Hy4-Flash Q3_K on 2x RTX 3090 with 128 GB DDR4 RAM, getting PP ~60-65 t/s and TG ~20-24 t/s. Setup is llama.cpp (llama-server router mode, a PR build) with Q3_K GGUF weights, q8_0 KV cache and 131072 context, using expert streaming via --moe-cache-mib 33768 and -cmoe, so experts run from system RAM. A logged run shows 4091 prompt tokens at 64.54 t/s and 100 generated tokens at 21.95 t/s. User says the model is slower but finishes answers with fewer tokens, and calls it a likely goto model.

Nov 18, 2026
Showing 1–5 of 5
Page 1 of 1

Community benchmarks snapshot

Records by GPU

AMD Strix Halo 128GB347NVIDIA RTX 3090263NVIDIA RTX 5090196NVIDIA DGX Spark97NVIDIA RTX 3060 12GB93NVIDIA RTX 5060 Ti 16GB86NVIDIA RTX Pro 6000 Blackwell81AMD Radeon AI PRO R9700 32GB77NVIDIA RTX 5070 Ti76M5 Max 48GB67

Records by model

3210 total
Qwen3.81553
Qwen3.6331
Gemma 4182
DeepSeek V4 Flash150
Qwen3.5143
DeepSeek V4.1 Flash102
other749

Records by engine

2275 total
llama.cpp1119
Strata231
vLLM221
MLX107
Ollama96
other501

Use cases

coding 923agentic 624long-context 428tool-use 269vision 165summarization 78math 75creative-writing 57multilingual 34rp 20text-generation 9reasoning 3
coding923agentic624long-context428tool-use269vision165summarization78math75creative-writing57

Median generation t/s by GPU

On Qwen3.8 27B at 4-bit, plain single-GPU runs. Full ranking

RTX 509080RTX Pro 600067RTX 5070 Ti55M5 Ultra 256GB50RTX 409040RX 7900 XTX40RTX 309037RX 7800 XT 16GB35RTX 5090 Laptop 24GB34RX 7900 XT32

Prompt processing, same model

M5 Ultra 256GB1800RTX 40901949RX 7900 XTX640RTX 3090561RX 7900 XT800M3 Ultra 512GB372V100 32GB941Radeon AI PRO R9700 32GB1287RTX 3080 20GB997Arc Pro B70512

Reports by model size

Qwen3.8 27B814Qwen3.8 125B · 6B active667Qwen3.6 35B · 3B active239DeepSeek V4 Flash 284B · 13B active133DeepSeek V4.1 Flash 552B · 16B active91Qwen3.6 27B81Gemma 4 26B · 4B active61GLM-5.3 320B · 18B active57

Quants

Q4_K_M390UD-Q4_K_XL185NVFP4170Q8_01194-bit106IQ4_XS85IQ3_S84Q4_K_XL83IQ3_XXS82Q480