PicoMistral 23M — Hardware Requirements & GPU Compatibility
Chat
PicoMistral 23M is a 24M-parameter open language model from PicoKittens in the Mistral family. It supports a context window of up to 512 tokens. At Q4_K_M it needs about 0.32 GB of VRAM — see which GPUs and Macs can run it below.
Runs great on a 24 GB GPU
Runs from 8 GB (Q4_K_M)Best on 24 GB (Q4_K_M · ~2048 tok/s)Macs from 8 GB (Q4_K_M · ~224 tok/s)
Apache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run PicoMistral 23M
One recommended way. Every other method, OS and quant is in the step-by-step guide.
Recommended
vllm serve PicoKittens/PicoMistral-23M --max-model-len 512vLLM repo with its own quantisation
Download: Weights on Hugging Face
Other ways
each opens step-by-stepRun in cloud
Fits on RTX 3060 12GB (11 GB headroom) · Q4_K_M
- Generation speed
- ~731 tok/s
- generation speed
- Cost per 1M output tokens
- $0.032
- per 1M output tokens
How Much VRAM Does PicoMistral 23M Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 0.3 GB | — | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: — context · ~2048 tok/s | |
| 0.3 GB | — | 3-bit medium — 3-bit medium quantization | Fits: — context · ~2048 tok/s | |
| 0.3 GB | — | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: — context · ~2048 tok/s | |
| 0.3 GB | — | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: — context · ~1986 tok/s | |
| 0.3 GB | — | 6-bit, very good — 6-bit quantization, very good quality | Fits: — context · ~1986 tok/s | |
| 0.3 GB | — | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: — context · ~1986 tok/s | |
| 0.4 GB | — | 16-bit · bfloat — Brain floating point 16 — preferred for training | Fits: — context · ~1820 tok/s |
Selected Q4_K_M
4-bit medium quantization0.3GBat 1K context
- Weights
- 0.1 GB
- KV cache
- 0.0 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 23.7 GB free · ~2048 tok/s
≥50 tok/s needs ≥25 GB/s · RTX 5090 ~3640, RTX 3090 Ti ~2048
Which Hardware Can Run PicoMistral 23M?
Q4_K_M · 0.3 GBPicoMistral 23M (Q4_K_M) requires 0.3 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 1+ GB is recommended. 52 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 163 devices with unified memory can run PicoMistral 23M, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB).
≤ 11 GB
≤ 11 GB tier →fits · 7.7 GB free · up to — ctx
16 GPUs · 1 Mac class · 2 device families
RTX 3080~1544Simulate output on RTX 3080RTX 2080 Ti~1251Simulate output on RTX 2080 TiRTX 3070 Ti~1236Simulate output on RTX 3070 TiM3*8 GB~224Simulate output on iMac 24" M3 (8 GB)Orin Nano~207Simulate output on NVIDIA Jetson Orin Nano 8GB (Super)iPhone~149Simulate output on iPhone 1720 GB
fits · 19.7 GB free · up to — ctx
1 GPU
RX 7900 XT~1500Simulate output on RX 7900 XT32–40 GB
fits · 31.7 GB free · up to — ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~3640Simulate output on RTX 5090M5 Max36 GB~1006Simulate output on Mac Studio M5 Max (36 GB)M532 GB~336Simulate output on MacBook Air 13" M5 (32 GB)Orin~416Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~274Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
Alternatives to PicoMistral 23M
Mistral family · 34 models →Similar models
1| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| The GuageLLM 23MHai929 | — | 0 GB | ~32760 | same size |
Specifications
- Publisher
- PicoKittens
- Family
- Mistral
- Parameters
- 24M
- Architecture
- MistralForCausalLM
- Context Length
- 512 tokens
- Vocabulary Size
- 16,384
- Release Date
- 2026-02-24
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does PicoMistral 23M need?
PicoMistral 23M requires 0.3 GB of VRAM at Q4_K_M, or 0.4 GB at BF16.
VRAM = Weights + KV Cache + Overhead
Weights = 24M × 4.8 bits ÷ 8 = 0 GB
KV Cache + Overhead ≈ 0.3 GB (at 2K context + ~0.3 GB framework)
Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.
VRAM usage by quantization
What's the best quantization for PicoMistral 23M?
For PicoMistral 23M, Q4_K_M (0.3 GB) offers the best balance of quality and VRAM usage. Q5_K_M (0.3 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 0.3 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run PicoMistral 23M on a Mac?
Yes — MacBook Air 13" M3 (8 GB) and 145 other Macs can run PicoMistral 23M. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 0.3 GB of usable unified memory (RAM minus macOS overhead).
Can I run PicoMistral 23M locally?
Yes — PicoMistral 23M can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is PicoMistral 23M?
At Q4_K_M, PicoMistral 23M can reach ~15000 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~2048 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: NVIDIA B200 → 8000 ÷ 0.32 × 0.65 = ~16250 tok/s
Estimated speed at Q4_K_M (0.3 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of PicoMistral 23M?
At Q4_K_M, the download is about 0.01 GB. The full-precision BF16 version is 0.05 GB. The smallest option (Q2_K) is 0.01 GB.
Which GPUs can run PicoMistral 23M?
52 consumer GPUs can run PicoMistral 23M at Q4_K_M. Top options include AMD Radeon RX 6700 XT, AMD Radeon RX 6800, AMD Radeon RX 6800 XT. 52 GPUs have plenty of headroom for comfortable inference.
Which devices can run PicoMistral 23M?
163 devices with unified memory can run PicoMistral 23M at Q4_K_M (0.3 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.