Qwen 9B vs 27B for local text
Text and reasoning with a smaller local setup
Up to three models side by side, with sources.
Q4 · each model's own figure, before any machine's reserve
No score in LiveBench 2026-06-25 for these models (not drawn, not zero) Not in LiveBench 2026-06-25
Q4 planning estimate: weights plus working room, before any machine's reserve; not a local measurement.
Same benchmark name on each model card, values as printed. Each publisher chose its own prompts, shots and harness, so these are not comparable and never ranked.
| Benchmark | Phi-4 | Mistral Small 3.2 24B |
|---|---|---|
| MMLU | 84.8 (simple-evals) | 80.50% |
| Specification | Phi-4 | Mistral Small 3.2 24B |
|---|---|---|
| Publisher | Microsoft | Mistral |
| License | MIT | Apache 2.0 |
| Task and abilities | Text & chat | Text & chatVision |
| Parameters | 14 B | 24 B |
| Memory needed · Q4 · model only | Needs ~11 GB | Needs ~19 GB |
| Disk · Q4 weights | ~9 GB | ~15 GB |
| Context · tokens | 16,384 | 131,072 |
| LiveBench rank · open models | No score in this LiveBench release | No score in this LiveBench release |
| Measured speed | No published tokens-per-second measurement for this checkpoint | 7 published rows |
| Source | Model card | Model card |
See which of these models run on your machine, well or slowly.
Written comparisons with the same evidence, sources and limits.
Text and reasoning with a smaller local setup
Coding quality under a 64 GB planning ceiling
Choosing an image editing or text-to-image workflow
The two most-searched 8B chat models for an 8 to 16 GB computer
Two mixture-of-experts chat models for a 16 to 32 GB machine
A 14B and a 24B dense model for a 16 to 24 GB machine
Two open coding models for a 24 to 48 GB machine
A newer and an older 8B model for an 8 to 16 GB computer
Two Apache 2.0 8B models for business use on a small machine
Two mixture-of-experts chat models for a 24 to 32 GB machine
Two dense ~30B models for a 24 to 48 GB machine
Two popular text-to-image models for a 12 to 24 GB graphics card
A very large and a mid-size open text-to-image model
Two small, fast text-to-image models for a 12 to 16 GB card
Two mixture-of-experts vision models that answer questions about images
Two small vision models that describe and read images on an 8 to 16 GB machine
Two open text-to-video models and the GPU memory they document
Local speech-to-text: the full model or the faster turbo
Two small multilingual embedding models for local search and RAG
A tiny and a mid-size multilingual embedding model