Qwen 9B vs 27B for local text
Text and reasoning with a smaller local setup
Facts from the reviewed catalog
| Specification | Qwen3.5 9B | Qwen3.8 27B |
|---|---|---|
| Task | Text & chat | Text & chat |
| Parameters | 9B | 27B |
| License | Apache 2.0 | Apache 2.0 |
| Commercial use | Allowedlicense text | Allowedlicense text |
| Quality · LiveBench 2026-06-25 | No score in LiveBench 2026-06-25 · source | 75.3overall · 0–100 · source |
| Memory | Q4 planning estimate (~7 GB of weights before the machine's reserve) | Q4 planning estimate (~21 GB of weights before the machine's reserve) |
| Checkpoint | official card | official card |
Only Qwen3.8 27B has a result in LiveBench 2026-06-25 (75.3); Qwen3.5 9B has none, so the two are not ranked from this source and no score is inferred for it. Questions with verifiable answers, released monthly; a score belongs to its release. Scale: overall score, 0–100, higher is better.
Memory on three machines
Each machine's own reserve is added (graphics cards 2 GB; unified and system memory 25%, 2 to 8 GB). Planning estimates, not measured runs.
| Machine | Qwen3.5 9B | Qwen3.8 27B |
|---|---|---|
| RTX 3060 · 12 GB VRAM | ~9 GBFast memory holds it | ~23 GBIn memory, outside the fast part · offload to PC RAM (32 GB assumed) |
| RTX 4090 · 24 GB VRAM | ~9 GBFast memory holds it | ~23 GBFast memory holds it |
| MacBook Air 13-inch M4 · 32 GB unified | ~15 GBFast memory holds it | ~29 GBFast memory holds it |
Qwen3.5 9B: Q4 planning estimate (~7 GB of weights before the machine's reserve). Qwen3.8 27B: Q4 planning estimate (~21 GB of weights before the machine's reserve). Formulas
Everything that runs on each machine: RTX 3060 · 12 GB VRAM · RTX 4090 · 24 GB VRAM · MacBook Air 13-inch M4 · 32 GB unified
Which to choose
Both are Apache 2.0 Qwen checkpoints that take text, image and video input. The 9B is a dense model whose Q4 weights stay under 7 GB, so a 12 GB graphics card holds it with room for context. The 27B is also dense: every token reads all 27 billion parameters, which needs roughly three times the memory and is slower per token on the same machine. Pick the 9B when memory is the limit; consider the 27B only on a 24 GB card or a Mac with 32 GB or more.
Files and run commands
The default file of the most downloaded collected source, or the commands each card documents. Not run by this site.
Qwen3.5 9B
unsloth/Qwen3.5-9B-GGUF · Qwen3.5-9B-Q4_K_M.gguf · Q4_K_M · 5.7 GB · unsloth
ollama run hf.co/unsloth/Qwen3.5-9B-GGUF:Q4_K_Mllama-server -hf unsloth/Qwen3.5-9B-GGUF:Q4_K_Mlms get https://huggingface.co/unsloth/Qwen3.5-9B-GGUF- mlx-lm: mlx-lm runs MLX repositories; pick the MLX version on a Mac
Qwen3.8 27B
unsloth/Qwen3.8-27B-GGUF · Qwen3.8-27B-UD-Q4_K_M.gguf · UD-Q4_K_M · 16.5 GB · unsloth
ollama run hf.co/unsloth/Qwen3.8-27B-GGUF:UD-Q4_K_Mllama-server -hf unsloth/Qwen3.8-27B-GGUF:UD-Q4_K_Mlms get https://huggingface.co/unsloth/Qwen3.8-27B-GGUF- mlx-lm: mlx-lm runs MLX repositories; pick the MLX version on a Mac
Numbers each publisher reports
Each card's own results with its own settings. Publishers choose different tests and settings, so these values are never compared across publishers here.
Qwen3.5 9B · reported by Alibaba
- MMLU-Pro
- 82.5Language
- MMLU-Redux
- 91.1Language
- C-Eval
- 88.2Language
- SuperGPQA
- 58.2Language
- GPQA Diamond
- 81.7Language
- IFEval
- 91.5Language
- IFBench
- 64.5Language
- MultiChallenge
- 54.5Language
Every reported result
- AA-LCR
- 63.0Language
- LongBench v2
- 55.2Language
- HMMT Feb 25
- 83.2Language
- HMMT Nov 25
- 82.9Language
- LiveCodeBench v6
- 65.6Language
- OJBench
- 29.2Language
- BFCL-V4
- 66.1Language
- TAU2-Bench
- 79.1Language
- VITA-Bench
- 29.8Language
- DeepPlanning
- 18.0Language
- MMMLU
- 81.2Language
- MMLU-ProX
- 76.3Language
- NOVA-63
- 55.9Language
- INCLUDE
- 75.6Language
- Global PIQA
- 83.2Language
- PolyMATH
- 57.3Language
- WMT24++
- 72.6Language
- MAXIFE
- 83.4Language
- MMMU
- 78.4Vision Language
- MMMU-Pro
- 70.1Vision Language
- MathVision
- 78.9Vision Language
- Mathvista (mini)
- 85.7Vision Language
- We-Math
- 75.2Vision Language
- DynaMath
- 83.6Vision Language
- ZEROBench
- 3.0Vision Language
- ZEROBench_sub
- 31.1Vision Language
- VlmsAreBlind
- 93.7Vision Language
- RealWorldQA
- 80.3Vision Language
- MMStar
- 79.7Vision Language
- MMBench EN-DEV-v1.1
- 90.1Vision Language
- SimpleVQA
- 51.2Vision Language
- HallusionBench
- 69.3Vision Language
- OmniDocBench1.5
- 87.7Vision Language
- CharXiv (RQ)
- 73.0Vision Language
- MMLongBench-Doc
- 57.7Vision Language
- CC-OCR
- 79.3Vision Language
- AI2D_TEST
- 90.2Vision Language
- OCRBench
- 89.2Vision Language
- ERQA
- 55.5Vision Language
- CountBench
- 97.2Vision Language
- RefCOCO (avg)
- 89.7Vision Language
- EmbSpatialBench
- 83.0Vision Language
- RefSpatialBench
- 58.5Vision Language
- LingoQA
- 80.4Vision Language
- Hypersim
- 13.5Vision Language
- Nuscene
- 11.8Vision Language
- VideoMME (w sub.)
- 84.5Vision Language
- VideoMME (w/o sub.)
- 78.4Vision Language
- VideoMMMU
- 78.9Vision Language
- MLVU
- 84.4Vision Language
- MVBench
- 74.4Vision Language
- LVBench
- 70.0Vision Language
- MMVU
- 67.8Vision Language
- ScreenSpot Pro
- 65.2Vision Language
- OSWorld-Verified
- 41.8Vision Language
- AndroidWorld
- 57.8Vision Language
- SLAKE
- 79.0Vision Language
- PMC-VQA
- 57.9Vision Language
- MedXpertQA-MM
- 49.9Vision Language
Values as printed on the card, with the publisher's own settings (card section: Benchmark Results › Language; Benchmark Results › Vision Language); column “Qwen3.5-9B”. Not comparable across publishers and not used in any ranking. Model card
Qwen3.8 27B · reported by Alibaba
- Agentic terminal coding Terminal Bench 2.1 (Terminus)
- 73.0Text Performance
- Agentic coding SWE-bench Pro
- 61.7Text Performance
- Repo-level code generation NL2Repo-Bench
- 42.3Text Performance
- Agentic coding DeepSWE 1.1
- 42.2Text Performance
- Software engineering QwenSWEBench
- 79.0Text Performance
- Long-horizon office work CoWorkBench
- 70.7Text Performance
- Professional job tasks JobBench
- 33.4Text Performance
- Instruction following IFBench
- 79.5Text Performance
Every reported result
- Scientific reasoning GPQA Diamond
- 89.2Text Performance
- Multidisciplinary reasoning HLE
- 30.8Text Performance
- Competitive coding LiveCodeBench v6
- 90.3Text Performance
- Computer use OSWorld-Verified
- 84.3VL Performance
- Browser use WebArena-Verified
- 64.8VL Performance
- Mobile use AndroidWorld
- 81.9VL Performance
- Application recreation RecreationBench
- 47.1VL Performance
- Multimodal software engineering SWE-MM
- 38.6VL Performance
- Visual web development Vision2Web
- 62.9VL Performance
- Document intelligence OmniDocBench 1.5
- 91.1VL Performance
- Real-world perception RealWorldQA
- 85.9VL Performance
- Embodied intelligence ERQA
- 65.5VL Performance
Values as printed on the card, with the publisher's own settings (card section: Benchmark Results › Text Performance; Benchmark Results › VL Performance); column “Qwen3.8-27B”. Not comparable across publishers and not used in any ranking. Model card
Limits and dates
The published quality benchmark does not test a local quantized build. Memory estimates are not speed measurements or guaranteed hardware fits. Repository size, when available, is not VRAM. Formulas
Page updated · Catalog reviewed · LiveBench 2026-06-25, collected