nace-ai / kapitsa
git clone https://github.com/nace-ai/kapitsa.git
v0.1.39 · 1d ago
Blazing fast Apple Silicon optimized inference engine built for Qwen and Nemotron. OpenAI + Anthropic API compatible. Built by folks from nace.ai
—
stars, 24h
+157
stars, 7 days · #341
+157
stars, 30 days · #1,552
157
★ total · +157% this week
25
forks · 6 open issues
Star history starts 2026-10-20; the chart appears after the second daily snapshot.
LanguageC++
LicenseApache-2.0
Created2026-10-16 (4d ago)
Last pushtoday
Watchers11
OwnerOrganization
Links
- GitHub: github.com/nace-ai/kapitsa
- Releases · Issues · Hacker News mentions
README excerpt
Quickstart · CLI · Benchmarks · Models · Docs ## Quickstart **Requirements:** a Mac with Apple silicon (M1 or newer) and macOS 14 Sonoma or later. [code] [code] On a 16-24 GB Mac use `--model 9b`, or `4b` for the quickest replies. ## CLI [code] `--model` picks the model and its precision (`35b:6bit`). `kapitsa --help` has the options. Usage guide ## Benchmarks Writing speed of a coding answer, each engine in its fastest setup. 🟩 fastest per model · — not measured · details ### MacBook Pro (16-inch, 2024) · M4 Pro · 48 GB #### Qwen | Model | **Kapitsa** | oMLX | mlx‑serve | Ollama | MTPLX | LM Studio | llama.cpp | MLX | Rapid‑MLX | | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | | Qwen3.5 4B 4-bit | 🟩 **179 tok/s** | 102 | 137 | 109 | 104 | 81 | 70 | 88 | 98 | | Qwen3.8 27B 4-bit | 🟩 **63 tok/s** | 45 | 54 | 32 | 36 | 15 | 18 | 28 | — | | Qwen3.6 35B‑A3B 4-bit | 🟩 **154 tok/s** | 127 | 124 | 89 | 80 | 86 | 69 | 86 | — | #### Nemotron | Model | **Kapitsa** | oMLX | llama.cpp | MLX | | --- | ---: | ---: | ---: | ---: | | Nemotron Dory1.0 4B NVFP4 | 🟩 **64 tok/s** | — no support | 6 | 48 | | Nemotron 3.5 Lightning 30B‑A3B NVFP4 | 🟩 **106 tok/s** | — | — | 65 | | Nemotron 3.5 Bolt 74B‑A9B 2/4/8-bit mix | 🟩 **64 tok/s** | 35 | — no Metal support | 32 | ### MacBook Pro (16-inch, 2026) · M5 Max · 128 GB #### Qwen | Model | **Kapitsa** | oMLX | mlx‑serve | Ollama | MTPLX | LM Studio | llama.cpp | MLX | Rapid‑MLX | | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | | Qwen3.5 4B 4-bit | 🟩 **332 tok/s** | 192 | 271 | 213 | 175 | 154 | 160 | 148 | 187 | | Qwen3.8 27B 4-bit | **102 tok/s** | 93 | 🟩 **130** | 68 | 69 | 32 | 51 | 54 | 61 | | Qwen3.6 35B‑A3B 4-bit | 🟩 **281 tok/s** | 232 | 264 | 166 | 123 | 127 | 171 | 132 | 98 | | Qwen3.8 Flash‑Next 180B‑A7B 4-bit | **105 tok/s** | 107 | 🟩 **108** | 60 | 88 | — won't load | 96 | 77 | — won't load | #### Nemotron | Model | **Kapitsa** | oMLX | mlx‑serve | Ollama | MTPLX | LM Studio | llama.cpp | MLX | | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | | Nemotron Dory1.0 4B NVFP4 | 🟩 **117 tok/s** | — no support | — no support | — no support | — no support | — no support | 9 | 76 | | Nemotron 3.5 Lightning 30B‑A3B NVFP4 | 🟩 **239 tok/s** | 190 | 190 | 153 | 131 | 110 | 159 | 139 | | Nemotron 3.5 Bolt 74B‑A9B 2/4/8-bit mix | 🟩 **120 tok/s** | 69 | — no support | — no support | — no support | — no support | — no Metal support | 69 | | Nemotron 3.5 Super‑VL 120B‑A12B NVFP4 | 🟩 **85 tok/s** | 52 | — no support | 64 | 44 | 61 | 66 | 46 | ### MacBook Pro (14-inch, 2021) · M1 Pro · 16 GB #### Qwen | Model | **Kapitsa** | oMLX | mlx‑serve | Ollama | MTPLX | LM Studio | llama.cpp | MLX | Rapid‑MLX | | --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: | | Qwen3.5 4B 4-bit | 🟩 **85 tok/s** | 55 | 62 | 55 | 42 | 47 | 38 | 46 | 52 | | Qwen3.5 9B 4-bit | 🟩 **51 tok/s** | — not run | 37 | — not run | — not run | — not run | — not r