DeepSeek R1 (8B)
Reasoning-tuned 8B model. Slower than a plain chat model, but thinks through its answers before replying.
At a glance
- Parameters
- 8B
- Quantization
- Q4_K_M
- Context window
- 128K tokens
- Approx. size
- 5.2 GB
- Engines
- llama.cpp, MLX, Ollama
- License
- MIT (open)
- Version
- v1.1.0 , released 2026-04-26
- Family
- deepseek
Tuning
Defaults a gezel client applies out of the box, from real evaluation runs against this exact quantization.
- sampling.temperature
0.6- sampling.topP
0.95- reasoning.thinkingBudget
512- reasoning.enableThinking
true
Model behaviors
Client-side behavior modules the catalog enables for this model (reasoning-tag handling, fabrication detection, prompt shaping).
- reasoning.strip-think-tags
- prompt.private-reasoning-guidance
- prompt.tool-cookbook-condensed
- fabrication.detect-past-tense-no-tools
- turn.preamble-folding
- turn.ramble-detection
Sources
- llama.cpp (GGUF)
- unsloth/DeepSeek-R1-0528-Qwen3-8B-GGUF ·
DeepSeek-R1-0528-Qwen3-8B-Q4_K_M.gguf· Q4_K_M - MLX
- mlx-community/DeepSeek-R1-0528-Qwen3-8B-4bit
- Ollama
deepseek-r1:8b- Upstream
- https://ollama.com/library/deepseek-r1
- Manifest
- View on GitHub
Tags
- deepseek
- tools
- reasoning