Gemma 4 26B-A4B
Gemma · Apache 2.0
Mixture-of-Experts architecture: 25.2B total parameters but only 3.8B active per token (8 selected + 1 shared expert per layer, out of 128 total). Hybrid dense+sparse FFN design. Inference throughput closer to a 4B dense model; quality closer to a 27B dense model. 256K context window. Benchmarks: 88.3% AIME 2026, 82.6% MMLU Pro, 77.1% LiveCodeBench. All 25.2B weights must be loaded into VRAM despite sparse activation; fits on 24 GB GPUs at Q4_K_M. Apache 2.0 licensed.
- Parameters
- 25.2B
- Architecture
- Dense
- Context
- 256,000 tokens
- Released
- 2026-04-02
- Engines
- llama.cpp, ollama, vLLM
- Builder Tools
- Claude Code, Codex CLI, Continue, Cursor, LM Studio, Ollama, Open WebUI, Windsurf
Parameters
25.2B
VRAM
24 GB
Context
250K
Formats
5
GPUs
37
Gemma 4 26B-A4B (25.2B) requires 24 GB VRAM at recommended quality (Q6_K). At efficient quality (Q4_K_M), it fits in 18 GB VRAM, making it compatible with the NVIDIA RTX 4090 Laptop (150-175W). On NVIDIA Grace Blackwell Ultra GB300, expect approximately 500 tok/s at Q8_0. For the best experience, AMD AI Powerhouse ($1,818) is recommended.
Source: OwnRig methodology
24 GB
Q6_K
22.86 GB
250K tokens
Chat
VRAM Requirements
| Quality | Quantization | VRAM | File Size |
|---|---|---|---|
| full | Q8_0 | 28 GB | 26.86 GB |
| recommended | Q6_K | 24 GB | 22.86 GB |
| recommended | Q5_K_M | 20.5 GB | 19.32 GB |
| efficient | Q4_K_M | 18 GB | 17.04 GB |
| compressed | Q3_K_M | 14 GB | 12.36 GB |
Context Length Impact
KV cache VRAM at Q6_K quality. Longer context = more memory.
| Context | KV Cache | Total VRAM |
|---|---|---|
| 2K | 205 MB | 24.2 GBexceeds 24 GB |
| 4K | 410 MB | 24.4 GBexceeds 24 GB |
| 8K | 819 MB | 24.8 GBexceeds 24 GB |
| 16K | 1.5 GB | 25.5 GBexceeds 24 GB |
| 32K | 3.1 GB | 27.1 GBexceeds 24 GB |
| 64K | 6.1 GB | 30.1 GBexceeds 24 GB |
| 128K | 12.3 GB | 36.3 GBexceeds 24 GB |
Compatible GPUs
37 devicesShowing 37 of 37 entries
Builder Context
Gemma 4 26B-A4B is commonly used with Claude Code, Codex CLI, Continue, Cursor, LM Studio, Ollama, Open WebUI, Windsurf. For an AI coding workflow, pair it with an embedding model like nomic-embed-text for local RAG.
Frequently Asked Questions
- How much VRAM does Gemma 4 26B-A4B need?
- Gemma 4 26B-A4B requires 24 GB VRAM at recommended quality (Q6_K). At lower quality settings, it can fit in as little as 14 GB.
- What is the best GPU for Gemma 4 26B-A4B?
- The NVIDIA Grace Blackwell Ultra GB300 delivers the best performance for Gemma 4 26B-A4B, achieving 500 tok/s at Q8_0 with an excellent rating.
- Can I run Gemma 4 26B-A4B on an RTX 4060 Ti?
- Yes. On the NVIDIA GeForce RTX 4060 Ti 16GB, Gemma 4 26B-A4B runs at 98 tok/s (Q3_K_M, excellent).
- What quantization should I use for Gemma 4 26B-A4B?
- For the best quality, use Q6_K (24 GB VRAM). If your GPU has limited VRAM, Q3_K_M (14 GB) is the most efficient option with acceptable quality.
- Is Gemma 4 26B-A4B good for coding?
- Yes. Gemma 4 26B-A4B is used with Claude Code, Codex CLI, Continue, Cursor, LM Studio, Ollama, Open WebUI, Windsurf for local AI coding. For the best coding experience, pair it with an embedding model for local RAG.
Related Guides
Tutorial
Running Gemma 4 locally: which GPU you actually need
Gemma 4 VRAM requirements for every variant: E2B, E4B, 26B-A4B, and 31B. Which GPUs can run each, what quantization to use, and the honest call on RTX 4060 vs RTX 4090.
Buying Guide
Mac Mini M4 for AI: which models run on 16 GB
Which AI models run on the Mac Mini M4 with 16 GB, 24 GB, or 48 GB of unified memory. Honest compatibility table, real quantization requirements, and the upgrade case for M4 Pro.
Data confidence: estimated. Source
VRAM requirements are calculated from model parameters and may vary by inference engine, context length, and batch size. Performance estimates are based on community benchmarks and should be verified for your specific configuration.Gemma is a trademark of its respective owner. OwnRig is not affiliated with or endorsed by the model creator.