Bielik — local AI model by SpeakLeash / ACK Cyfronet AGH
Written by Jakub Rusinowski · Last updated
Poland's open LLM family from SpeakLeash and ACK Cyfronet AGH, trained on Polish large-scale computing infrastructure (PLGrid) on a corpus curated by the SpeakLeash team. The v3 line runs from 11B through the pruned Minitron 7B down to 4.5B and 1.5B. The 11B and Minitron 7B come in two builds: one with a multilingual tokenizer, and a PL build with SpeakLeash's APT4 tokenizer, which encodes Polish in fewer tokens and so does more with its parameters on Polish text. Bielik 11B v3.1 (October 2026) adds reasoning and tool calling. SpeakLeash also publishes Bielik Guard, a small Polish safety classifier (0.1B and 0.5B) that scores text for hate, vulgarity, sexual content, crime and self-harm; it is a classifier, not a chat model, so it has no page here.
Variants
The smallest Bielik variant needs about 2 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache.
| Model | VRAM at Q4 | VRAM | Context | Run it |
|---|---|---|---|---|
| Bielik PL 11B v3.0 Instruct → 11B | ~7.4 GB | 32,768 | ollama run bielik | |
| Bielik 11B v3.1 Instruct → 11B | ~7.5 GB | 32,768 | cloud-hosted tag | |
| Bielik 11B v3.0 (multilingual) → 11B | ~7.5 GB | 32,768 | ollama run bielik | |
| Bielik Minitron 7B v3.0 Instruct → 7B | ~5.3 GB | 32,768 | ollama run bielik | |
| Bielik PL Minitron 7B v3.0 Instruct → 7B | ~5.3 GB | 32,768 | ollama run bielik | |
| Bielik 4.5B v3.0 Instruct → 4.5B | ~3.7 GB | 8,192 | ollama run bielik | |
| Bielik 1.5B v3.0 Instruct → 1.5B | ~1.8 GB | 8,192 | ollama run bielik |
Memory is quantized weights plus overhead at Q4_K_M, from the same engine as the GPU & VRAM checker.
How to run Bielik locally
Install Ollama, then pull the tag.
ollama run bielikPick a size above for its own VRAM figure, speed estimate and install command.
Licence
Commercial use permitted. No usage restrictions beyond attribution.
Applies to: Bielik PL 11B v3.0 Instruct, Bielik 11B v3.1 Instruct, Bielik 11B v3.0 (multilingual), Bielik Minitron 7B v3.0 Instruct, Bielik PL Minitron 7B v3.0 Instruct, Bielik 4.5B v3.0 Instruct, Bielik 1.5B v3.0 InstructRecommended GPU
The cheapest catalogued GPU that runs Bielik locally (min 2 GB VRAM) is the Intel Arc B570 (10 GB).
Bielik — frequently asked questions
How much VRAM does Bielik need?
Bielik needs about 2 GB VRAM at Q4_K_M quantization for its smallest variant. Variants: Bielik PL 11B v3.0 Instruct (7 GB, Q4_K_M); Bielik 11B v3.1 Instruct (8 GB, Q4_K_M); Bielik 11B v3.0 (multilingual) (8 GB, Q4_K_M); Bielik Minitron 7B v3.0 Instruct (5 GB, Q4_K_M). On Apple Silicon, unified memory counts toward this requirement.
Can I run Bielik on an RTX 4090 (24 GB)?
Yes — Bielik runs on an RTX 4090 (24 GB) and other 24 GB cards such as the RTX 3090. Smaller variants also fit comfortably on 8–16 GB GPUs at Q4_K_M.
What quantization should I use for Bielik?
Q4_K_M is the best balance of quality and VRAM for Bielik in most cases. Choose Q8_0 for near-lossless quality if you have spare VRAM, or smaller quants (Q3/Q2) only when memory is tight.
How do I run Bielik with Ollama?
Bielik has no local Ollama tag — the published tag is cloud-hosted, so running it sends your prompts to a hosted GPU rather than your own machine.