Gemma 4 26B A4B vs Qwen3 30B A3B for local chat
Two mixture-of-experts chat models for a 24 to 32 GB machine
Facts from the reviewed catalog
| Specification | Gemma 4 26B A4B | Qwen3 30B A3B Instruct 2507 |
|---|---|---|
| Task | Text & chat | Text & chat |
| Parameters | 25.2B · 3.8B active | 30.5B · 3.3B active |
| License | Apache 2.0 | Apache 2.0 |
| Commercial use | Allowedlicense text | Allowedlicense text |
| Quality · LMArena 2026-10-22 · text | 1,437 ± 71,430–1,445 · 6,092 votes · rank 100 · source | 1,382 ± 51,377–1,387 · 23,306 votes · rank 190 · source |
| Memory | Q4 planning estimate (~19 GB of weights before the machine's reserve) | Q4 planning estimate (~23 GB of weights before the machine's reserve) |
| Checkpoint | official card | official card |
In LMArena 2026-10-22 · text, Gemma 4 26B A4B rates 1,437 ± 7 against 1,382 ± 5 for Qwen3 30B A3B Instruct 2507, 55 points ahead with intervals that do not overlap. Blind pairwise votes by people on the hosted model, CC BY 4.0. A rating is relative; overlapping intervals mean the order is not settled. Scale: Arena rating (style control, overall) ± half the confidence interval.
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 | Gemma 4 26B A4B | Qwen3 30B A3B Instruct 2507 |
|---|---|---|
| RTX 4090 · 24 GB VRAM | ~21 GBFast memory holds it | ~25 GBIn memory, outside the fast part · offload to PC RAM (32 GB assumed) |
| RTX 5090 · 32 GB VRAM | ~21 GBFast memory holds it | ~25 GBFast memory holds it |
| MacBook Air 13-inch M4 · 32 GB unified | ~27 GBFast memory holds it | ~31 GBFast memory holds it |
Gemma 4 26B A4B: Q4 planning estimate (~19 GB of weights before the machine's reserve). Qwen3 30B A3B Instruct 2507: Q4 planning estimate (~23 GB of weights before the machine's reserve). Formulas
Everything that runs on each machine: RTX 4090 · 24 GB VRAM · RTX 5090 · 32 GB VRAM · MacBook Air 13-inch M4 · 32 GB unified
Which to choose
Both activate about 4 billion parameters per token or fewer (3.8 billion for Gemma 4, 3.3 billion for Qwen3), so both generate quickly for their size, and both are Apache 2.0 with a 262,144-token context. Gemma 4 26B A4B is a little smaller in total (25.2 billion against 30.5), so its Q4 weights take about 19 GB against 23 GB, and it reads images; Qwen3 30B A3B Instruct 2507 is text only. In the same LMArena text release Gemma 4 26B A4B rates higher.
Files and run commands
The default file of the most downloaded collected source, or the commands each card documents. Not run by this site.
Gemma 4 26B A4B
unsloth/gemma-4-26B-A4B-it-GGUF · gemma-4-26B-A4B-it-UD-Q4_K_M.gguf · UD-Q4_K_M · 16.9 GB · unsloth
ollama run hf.co/unsloth/gemma-4-26B-A4B-it-GGUF:UD-Q4_K_Mllama-server -hf unsloth/gemma-4-26B-A4B-it-GGUF:UD-Q4_K_Mlms get https://huggingface.co/unsloth/gemma-4-26B-A4B-it-GGUF- mlx-lm: mlx-lm runs MLX repositories; pick the MLX version on a Mac
Qwen3 30B A3B Instruct 2507
unsloth/Qwen3-30B-A3B-Instruct-2507-GGUF · Qwen3-30B-A3B-Instruct-2507-Q4_K_M.gguf · Q4_K_M · 18.6 GB · unsloth
ollama run hf.co/unsloth/Qwen3-30B-A3B-Instruct-2507-GGUF:Q4_K_Mllama-server -hf unsloth/Qwen3-30B-A3B-Instruct-2507-GGUF:Q4_K_Mlms get https://huggingface.co/unsloth/Qwen3-30B-A3B-Instruct-2507-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.
Gemma 4 26B A4B · reported by Google
- MMLU Pro
- 82.6%
- AIME 2026 no tools
- 88.3%
- LiveCodeBench v6
- 77.1%
- Codeforces ELO
- 1,718
- GPQA Diamond
- 82.3%
- Tau2 (average over 3)
- 68.2%
- HLE no tools
- 8.7%
- HLE with search
- 17.2%
Every reported result
Values as printed on the card, with the publisher's own settings (card section: Benchmark Results); column “Gemma 4 26B A4B”. Not comparable across publishers and not used in any ranking. Model card
Qwen3 30B A3B Instruct 2507 · reported by Alibaba
- MMLU-Pro
- 78.4Knowledge
- MMLU-Redux
- 89.3Knowledge
- GPQA
- 70.4Knowledge
- SuperGPQA
- 53.4Knowledge
- AIME25
- 61.3Reasoning
- HMMT25
- 43.0Reasoning
- ZebraLogic
- 90.0Reasoning
- LiveBench 20241125
- 69.0Reasoning
Every reported result
- LiveCodeBench v6 (25.02-25.05)
- 43.2Coding
- MultiPL-E
- 83.8Coding
- Aider-Polyglot
- 35.6Coding
- IFEval
- 84.7Alignment
- Arena-Hard v2
- 69.0Alignment
- Creative Writing v3
- 86.0Alignment
- WritingBench
- 85.5Alignment
- BFCL-v3
- 65.1Agent
- TAU1-Retail
- 59.1Agent
- TAU1-Airline
- 40.0Agent
- TAU2-Retail
- 57.0Agent
- TAU2-Airline
- 38.0Agent
- TAU2-Telecom
- 12.3Agent
- MultiIF
- 67.9Multilingualism
- MMLU-ProX
- 72.0Multilingualism
- INCLUDE
- 71.9Multilingualism
- PolyMATH
- 43.1Multilingualism
Values as printed on the card, with the publisher's own settings (card section: Performance); column “Qwen3-30B-A3B-Instruct-2507”. 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 · LMArena 2026-10-22 · text