Gemma 4 E4B
- In Input modalities: TextImageAudio Out Output modalities: Text
- Reasoning
- Tool use
- Open weights
Prices
0 offerings · prices as listedNo priced offering is listed for this model yet.
Details
- Licence
- apache-2.0
- Family
- gemma
Not stated on Google’s pages: max input, reasoning efforts, reasoning can be off, max thinking budget, structured output, built-in tools, knowledge cutoff, aliases.
Run locally
Weights: google/gemma-4-E4B- Parameters 8B
- Dense
- Native precision BF16
- Layers 42
- KV cache per token 84 KB
- Config positions 128K (config.json max_position_embeddings)
- API context 125K (maker’s API)
- Needs
- ≈ 15.8 GB (weights 14.9 (maker’s file) + KV cache 0.1 + overhead 0.7)
- Usable
- 21.6 GB (90% of its memory)
- Generation speed
- 31–47 tok/s
- Longest context
- 125K (the model’s limit)
- First token
- about 0.26–0.39 s for a 2K prompt
| Precision | Weights | Needs at 8K | On your hardware | Longest context |
|---|---|---|---|---|
| BF16 (native)maker’s file | 14.9 GB | ≈ 15.8 | ●Runs well 31–47 tok/s | 125K |
| Q8_0 | 7.9 GB | ≈ 8.6 | ●Runs well 58–87 tok/s | 125K |
| FP8 | 7.4 GB | ≈ 8.1 | ●Runs well 62–93 tok/s | 125K |
| Q6_K | 6.1 GB | ≈ 6.8 | ●Runs well 75–112 tok/s | 125K |
| Q5_K_M | 5.3 GB | ≈ 5.9 | ●Runs well 86–129 tok/s | 125K |
| Q5_0 | 5.1 GB | ≈ 5.8 | ●Runs well 89–134 tok/s | 125K |
| Q4_K_M | 4.5 GB | ≈ 5.2 | ●Runs well 101–151 tok/s | 125K |
| Q4_0 | 4.2 GB | ≈ 4.8 | ●Runs well 108–163 tok/s | 125K |
| IQ4_XS | 4.0 GB | ≈ 4.6 | ●Runs well 114–172 tok/s | 125K |
| Q3_K_Mquality drops | 3.6 GB | ≈ 4.3 | ●Runs well 124–186 tok/s | 125K |
| IQ3_XXSquality drops | 2.8 GB | ≈ 3.5 | ●Runs well 157–235 tok/s | 125K |
| Q2_Kquality drops | 2.8 GB | ≈ 3.4 | ●Runs well 160–240 tok/s | 125K |
| IQ2_XXSquality drops | 1.9 GB | ≈ 2.6 | ●Runs well 228–342 tok/s | 125K |
Click a row to choose a precision. “Maker’s file” sizes are the maker’s own files; others are parameters × bits per weight.
Compared with other hardware
Each at the highest precision it runs. Click one for details.
Smallest that runs it (Q4_K_M, 8K): GPU GeForce RTX 3080; unified memory Apple M5 · 16 GB.
Run commands
vllm serve google/gemma-4-E4B --max-model-len 8192
Memory to fine-tune
Batch 1, 2,048-token sequences, gradient checkpointing; LoRA and QLoRA train rank-16 adapters on the attention projections.
Estimates; facts from the maker’s Hugging Face repositories, as of 2026-11-06. How it’s estimated
Cited scores
From third-party leaderboards, not measured by usThis model isn’t on any leaderboard we cite yet.
Changes
All changes →- New language model
- New language model
Sources
- Gemma 4 model card (opens in a new tab) ai.google.dev · checked Nov 1, 2026
- Release notes | Gemini API | Google AI for Developers (opens in a new tab) ai.google.dev · checked Nov 4, 2026
- google/gemma-4-E4B model card (opens in a new tab) huggingface.co · checked Nov 1, 2026
- google/gemma-4-E4B config.json (opens in a new tab) huggingface.co · checked Oct 30, 2026
- google/gemma-4-E4B files (opens in a new tab) huggingface.co · checked Oct 30, 2026
- Living Models pairs Gemma 4 with BOTANIC-1 to help decode plant DNA — Google DeepMind (opens in a new tab) deepmind.google · checked Nov 5, 2026
- google/gemma-4-E4B model card (opens in a new tab) huggingface.co · checked Nov 4, 2026
- google/gemma-4-E2B · Hugging Face (opens in a new tab) huggingface.co · checked Oct 30, 2026
Google’s official policies: data use, privacy, terms →
Some values on this page are still waiting to be checked against official pages: a filled dot is verified, a hollow one is not yet (with or without a source on file). How we verify data
Google · United States · ai.google.dev
Related models
Snapshots, the same family and other models from this provider