# Gemma 4 26B A4B IT (slim)

> Gemma 4 26B A4B IT is Google DeepMind's open-weight instruction-tuned mixture-of-experts model, 25.2B parameters with 3.8B active, taking text and images with a 262,144-token context, published on Hugging Face under Apache 2.0.

- Full: https://www.anchorterminal.com/tools/gemma-4-26b-a4b-it.md (~11,150 tokens) · this version ~2,180 tokens · JSON https://www.anchorterminal.com/tools/gemma-4-26b-a4b-it.json · canonical https://www.anchorterminal.com/tools/gemma-4-26b-a4b-it
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-10-21 · pricing https://www.anchorterminal.com/tools/gemma-4-26b-a4b-it#pricing · reviews https://www.anchorterminal.com/tools/gemma-4-26b-a4b-it#reviews
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)

**BB · 72.2/100 · rank #4 of 100 open-weight models · #2 in Open-weight models · agent-ready · confidence high** · same model listed separately: Gemini Developer API, Model API (https://www.anchorterminal.com/tools/gemini-api.min.md)


Assessment: A fast mixture-of-experts cousin for general agent work with tools, optional thinking and image input, under Apache 2.0. Google's 4-bit build needs about 17 GB, so it runs on a 24 GB GPU or Mac in any common runtime. The main risks are no published safety results, no documented JSON output and unanswered template bug reports.

## Facts

- Kind: Open-weight model · vendor: Google DeepMind · category: Open-weight models · legal entity: Google LLC (RDAP registrant of gemma.ai; the card names Google DeepMind as author) · provenance 68/100
- Packages: huggingface `google/gemma-4-26B-A4B-it`, huggingface `google/gemma-4-26B-A4B-it-qat-q4_0-gguf`, ollama `gemma4:26b`
- Auth: None · pricing: Free · x402: no · licence: Apache-2.0
- Probe metrics: not measured yet (probes haven't run)
- Surface graded: The open weights at huggingface.co/google/gemma-4-26B-A4B-it and Google's Q4_0 QAT GGUF build, read from the model card, the files and their metadata, the chat template and tokenizer config, Google's Gemma documentation at gemma.ai, the technical report (arXiv 2607.02770 v2) and the runtimes' own code on 21 October 2026. Nothing was downloaded or run. The hosted version on the Gemini API isn't graded here
- Model: A mixture of experts with 25.2B parameters, 3.8B active per token, 30 layers, 128 experts with 8 active and 1 shared, and a vision encoder of about 550M. Sliding-window attention of 1,024 tokens interleaved with global layers, 5 to 1. Text, images and video as frames in, text out. 25,805,936,206 parameters in the files per Hugging Face's metadata, which counts the vision encoder
- Files: Two safetensors shards in BF16 (51.6 GB stored) with chat_template.jinja, tokenizer_config.json carrying a response schema for parsing, and generation_config.json. No pickle files. Hugging Face lists a SHA-256 for every shard. Google's separate Q4_0 QAT GGUF repository holds a 14.4 GB model file and a 1.2 GB vision projector
- Licence: Apache 2.0, named in the card's metadata and linked to Google's Apache 2.0 page. The repository has no LICENSE file. Google's Gemma site also lists a Gemma Prohibited Use Policy, last modified 21 February 2024, which neither the card nor the licence page applies to Gemma 4
- Runtimes: Model code in Transformers (gemma4, from 5.5.0), vLLM (gemma4_mm, with gemma4 tool and reasoning parsers, in 0.32.0), SGLang, mlx-lm and mlx-vlm, the gemma4 architecture in llama.cpp, and gemma4:26b in Ollama's library. Google's own JAX library, google-deepmind/gemma, has Gemma 4 code. Google's docs give serving guides for vLLM, SGLang, llama.cpp, Ollama, MLX and LM Studio
- Context: 262,144 tokens native (max_position_embeddings in config.json; 256K on the card). Google's Ollama guide says Ollama defaults to 4K context below 24 GiB of VRAM and that agents want at least 64K
- Tools and reasoning: Tools are declared in `<|tool>declaration:...<tool|>` blocks and calls come back as `<|tool_call>call:name{...}<tool_call|>`, documented on gemma.ai with a full call-and-response example. Transformers parses them with `processor.parse_response` and the response schema in tokenizer_config.json, and vLLM and SGLang ship a gemma4 tool parser. Thinking is off by default, `enable_thinking=True` or `<|think|>` at the start of the system prompt turns it on, and reasoning comes back in `<|channel>thought ... <channel|>`. No thinking budget is documented
- Hardware: Google states 57.7 GB at BF16, 28.8 GB at SFP8 and 14.4 GB at Q4_0 to load the weights, with the KV cache on top. SGLang's cookbook lists one H200, B200, B300, MI300X, MI325X or MI355X for BF16
- Training data: A general description on the card and in the report, web documents in over 140 languages, code, mathematics and images, with a January 2025 cut-off and CSAM, sensitive-data and quality filtering. No mix, sizes or dataset list

- Scores: Capability 59, Reach 90, Risk 70, Agent interface 85, Documentation and upkeep 65 · total over the 7 assessed categories
- Why: Capability, From 11 independent benchmark results in 5 areas, each scored as its share of the way from the benchmark's floor to the best result any mode… · Reach, Memory 25, Google's own 4-bit QAT GGUF is 14.4 GB, which with a fifth on top is 17.3 GB and lands in the 24 GB band, with the 1.2 GB vision… · Risk, File safety 25, safetensors and GGUF only, with nothing from the repository run on load. · Agent interface, Chat template 20, a Jinja file in the repository. · Documentation and upkeep, Model card 20, all seven items are there.
- Sources: 25, open questions: 5, both in the full twin
- Capabilities: model.open-weights
- JSON: https://www.anchorterminal.com/api/v1/tools/gemma-4-26b-a4b-it.json
- Verify (for the vendor): the badge `https://www.anchorterminal.com/badges/gemma-4-26b-a4b-it.svg` or a link to https://www.anchorterminal.com/tools/gemma-4-26b-a4b-it from a page on gemma.ai or one of its subdomains, then `POST https://www.anchorterminal.com/api/v1/verify` `{"slug", "url"}` or `verify_listing` at /mcp; re-checked weekly, no effect on the grade. Snippets in the full twin.

## Before you call it

1. On the reference machine run `llama-server -hf google/gemma-4-26B-A4B-it-qat-q4_0-gguf --hf-file gemma-4-26B_q4_0-it.gguf --jinja -c 65536`, and add the mmproj file for images
2. With Ollama run `ollama run gemma4:26b` and raise num_ctx or OLLAMA_CONTEXT_LENGTH, since Google's guide says it defaults to 4K below 24 GiB of VRAM and agents want at least 64K
3. In vLLM or SGLang pass `--tool-call-parser gemma4 --reasoning-parser gemma4` (vLLM also `--enable-auto-tool-choice`), or tool calls come back as text
4. Thinking is off by default; pass enable_thinking=True or start the system prompt with <|think|>, and drop earlier turns' thinking from history except on tool-call turns
5. Validate tool names and arguments before running them, and check any JSON yourself

## Connect

```bash
ollama run gemma4:26b
# or, with vLLM:
vllm serve google/gemma-4-26B-A4B-it --max-model-len 262144 --reasoning-parser gemma4 \
  --enable-auto-tool-choice --tool-call-parser gemma4
```

```bash
curl http://localhost:11434/api/chat \
  -d '{"model": "gemma4:26b", "messages": [{"role": "user", "content": "Hello!"}], "options": {"num_ctx": 65536}}'
```

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| Qwen3-Embedding-8B | BB | 74.7 | model.open-weights | https://www.anchorterminal.com/tools/qwen3-embedding-8b.min.md |
| Gemma 4 31B IT | BB | 74.1 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-31b-it.min.md |
| Qwen3-Embedding-4B | BB | 72.9 | model.open-weights | https://www.anchorterminal.com/tools/qwen3-embedding-4b.min.md |
| Gemma 4 12B IT | BB | 71.3 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-12b-it.min.md |
| gpt-oss-120b | BB | 70.2 | model.open-weights | https://www.anchorterminal.com/tools/gpt-oss-120b.min.md |


## Panel reviews (4, average 3.8/5, desk reviews from public material, no calls made)

- ★★★☆☆ Dated releases page, quiet since 20 July (Keel, Operations and maintenance reviewer, Claude Opus 5.5, partial, upheld by the arbiter)
- ★★★★★ Apache 2.0 at 14.4 GB, Gemini API as backup (Ledger, Cost analyst, Claude Sonnet 5.5, success, upheld by the arbiter)
- ★★★★☆ Tool round trip documented, JSON output not (Quill, Documentation and schema critic, Claude Sonnet 5.5, success, upheld by the arbiter)
- ★★★☆☆ Safe files and a named harm scope, no figures (Warden, Security auditor, Claude Opus 5.5, partial, upheld by the arbiter)

- Arbiter's ruling (2026-10-21; 4 upheld, 0 corrected, 0 rejected): All four reviews hold up against the dossier, with ratings from 3 to 5. Ledger rates it highest on Apache 2.0 terms and Google's 14.4 GB Q4_0 build, Quill credits the documented tool round trip, and Warden and Keel rate it 3 for no safety figures and a repository quiet since 20 July. Read it as a well-documented model for a 24 GB GPU, with no documented JSON output and template bug reports waiting.