Gemma 3 4B
Googlehuggingface.co/google/gemma-3-4b-it
Gemma 3 introduces multimodality, supporting vision-language input and text outputs. It handles context windows up to 128k tokens, understands over 140 languages, and offers improved math, reasoning, and chat capabilities,...
Updated 2 min ago · first seen 11 Sept 2026
model_01M294WWJ4RN6BJMMMPP3G4MK3
- Parameters
- 4.3B
- T2 · 2 h ago
- Context
- 128K tokens
- T2 · 28 min ago
- Released
- 20 Feb 2025
- T2 · 2 min ago
- License
- gemma
- T2 · 2 h ago
- Knowledge cutoff
- Aug 2024
- T2 · 2 h ago
Specification
- Release date
- 20 Feb 2025
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 min agomedium
- Status
- deprecated
Source:Artificial AnalysisT2observed 28 min agomedium
- Openness
- restricted
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 18 min agomedium
- License
- gemma
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
- Architecture
- Gemma3ForConditionalGeneration
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
- Parameters
- 4.3B
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
- Context window
- 128K tokens
Source:Artificial AnalysisT2observed 28 min agomedium
- Max output
- 16.4K tokens
Source:OpenRouter public model & pricing listingT2observed 2 h agomedium
- Knowledge cutoff
- Aug 2024
Source:OpenRouter public model & pricing listingT2observed 2 h agomedium
- Modalities
- image, text
Source:OpenRouter public model & pricing listingT2observed 2 h agomedium
- Input modalities
- image, text
Source:OpenRouter public model & pricing listingT2observed 2 h agomedium
- Output modalities
- text
Source:OpenRouter public model & pricing listingT2observed 2 h agomedium
- Tokenizer
- Gemini
Source:OpenRouter public model & pricing listingT2observed 2 h agomedium
- Base model
- google/gemma-3-4b-pt
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
- File size
- 8.6 GB
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
- Hugging Face repo
- google/gemma-3-4b-it
Source:OpenRouter public model & pricing listingT2observed 2 h agomedium
- Pipeline tag
- image-text-to-text
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
- Model card
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
- Downloads
- 1,682,782
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 min agomedium
- Likes
- 1,483
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 min agomedium
- Gated
- manual
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
- Hf inference providers
- deepinfra, featherless-ai
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
- Instruct type
- gemma
Source:OpenRouter public model & pricing listingT2observed 2 h agomedium
- Last modified
- 2025-03-21T20:20:53+00:00
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
- Library name
- transformers
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
- Downloads all time
- 25,853,078
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 min agomedium
- Model type
- gemma3
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
- Openrouter id
- google/gemma-3-4b-it
Source:OpenRouter public model & pricing listingT2observed 2 h agomedium
- Supported parameters
- frequency_penalty, logit_bias, max_tokens, min_p, presence_penalty, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, top_k, top_p
Source:OpenRouter public model & pricing listingT2observed 2 h agomedium
- Tags
- transformers, safetensors, gemma3, image-text-to-text
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
- Weights dtype
- BF16
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium
Each value shows its source, tier and observation time. Conflicting claims are kept side by side and flagged — never averaged. How AI Atlas records facts →
Provenance
Attributed facts
36
Source tiers
T236
Freshest observation
2 min ago
Conflicts
None
Modalities
- Modalities
- imagetext
- Input
- imagetext
- Output
- text
Capabilities
Tool calling
No
OpenRouter public model & pricing listing · T2
Structured output
Yes
OpenRouter public model & pricing listing · T2
Reasoning
No
Artificial Analysis · T2
Vision
Yes
OpenRouter public model & pricing listing · T2
Audio
Unavailable
Fine-tuning available
Unavailable
- Context window
- 128K tokens
Source:Artificial AnalysisT2observed 28 min agomedium
- Max output
- 16.4K tokens
Source:OpenRouter public model & pricing listingT2observed 2 h agomedium
- Knowledge cutoff
- Aug 2024
Source:OpenRouter public model & pricing listingT2observed 2 h agomedium
- Tokenizer
- Gemini
Source:OpenRouter public model & pricing listingT2observed 2 h agomedium
| Benchmark | Score | Metric | Config | Evaluated | Source |
|---|---|---|---|---|---|
| τ²-bench | 4.97% | pass^1 | aa_slug=gemma-3-4b · evaluator=Artificial Analysis · index_version=4.3 | — | artificialanalysis.aiT2 |
| Artificial Analysis Intelligence Index | 4.83 | index | aa_slug=gemma-3-4b · version=4.3 · estimated=true | — | artificialanalysis.aiT2 |
| GPQA | 29.09% | accuracy | aa_slug=gemma-3-4b · variant=GPQA Diamond · evaluator=Artificial Analysis · index_version=4.3 | — | artificialanalysis.aiT2 |
| Humanity's Last Exam | 5.29% | accuracy | aa_slug=gemma-3-4b · evaluator=Artificial Analysis · index_version=4.3 | — | artificialanalysis.aiT2 |
| IFBench | 28.3% | accuracy | aa_slug=gemma-3-4b · evaluator=Artificial Analysis · index_version=4.3 | — | artificialanalysis.aiT2 |
| MMMU-Pro | 29.94% | accuracy | aa_slug=gemma-3-4b · evaluator=Artificial Analysis · index_version=4.3 | — | artificialanalysis.aiT2 |
| Terminal-Bench | 0.37% | accuracy | aa_slug=gemma-3-4b · variant=v2.1 · evaluator=Artificial Analysis · index_version=4.3 | — | artificialanalysis.aiT2 |
| Terminal-Bench | 0.76% | accuracy | aa_slug=gemma-3-4b · variant=hard · evaluator=Artificial Analysis · index_version=4.3 | — | artificialanalysis.aiT2 |
Scores are reported as published, with their evaluation configuration (harness, prompting, judge). Results with different configs are not directly comparable — see methodology.
Current prices
| Provider | Input / 1M | Output / 1M | Cached in | Batch in / out | Context | Observed | Source |
|---|---|---|---|---|---|---|---|
| Google Gemini APIgoogle/gemma-3-4b-it | $0.05 | $0.10 | — | — | 131.1K | 32 min ago | openrouter.aiT2 |
USD per 1M tokens as published by each provider (USD). Rows are append-only: every change is kept in the history below.
Available through 1
Price history
Input price · USD / 1M tokens
Output price · USD / 1M tokens
- Google Gemini API
Memory need = bytes per parameter (4-bit ≈ 0.5 × 1.15 overhead, 8-bit 1.0, fp16 2.0) + a KV-cache allowance. Not a measurement.
| Hardware | Quantization | Memory | Est. need | Fits |
|---|---|---|---|---|
| NVIDIA DGX B200 | 4bit | 1,440 GB | 3.0 GB | Yes |
| Apple M3 Ultra | 4bit | — | 3.0 GB | Yes |
| Mac Studio (Apple M5 Ultra) | 4bit | — | 3.0 GB | Yes |
| AMD Instinct MI325X | 4bit | 256 GB | 3.0 GB | Yes |
| AMD Instinct MI300X | 4bit | 192 GB | 3.0 GB | Yes |
| Apple M2 Ultra | 4bit | — | 3.0 GB | Yes |
| NVIDIA B200 | 4bit | 180 GB | 3.0 GB | Yes |
| NVIDIA H200 | 4bit | 141 GB | 3.0 GB | Yes |
| NVIDIA H200 NVL | 4bit | 141 GB | 3.0 GB | Yes |
| Apple M1 Ultra | 4bit | — | 3.0 GB | Yes |
| Apple M3 Max | 4bit | — | 3.0 GB | Yes |
| Apple M4 Max | 4bit | — | 3.0 GB | Yes |
| Mac Studio (Apple M5 Max) | 4bit | — | 3.0 GB | Yes |
| MacBook Pro (Apple M5 Max) | 4bit | — | 3.0 GB | Yes |
| NVIDIA DGX Spark | 4bit | 128 GB | 3.0 GB | Yes |
| Apple M2 Max | 4bit | — | 3.0 GB | Yes |
| NVIDIA H100 NVL | 4bit | 94 GB | 3.0 GB | Yes |
| NVIDIA A100 80GB | 4bit | 80 GB | 3.0 GB | Yes |
| NVIDIA H100 SXM | 4bit | 80 GB | 3.0 GB | Yes |
| Apple M1 Max | 4bit | — | 3.0 GB | Yes |
| Apple M4 Pro | 4bit | — | 3.0 GB | Yes |
| Mac mini (Apple M5 Pro) | 4bit | — | 3.0 GB | Yes |
| MacBook Pro (Apple M5 Pro) | 4bit | — | 3.0 GB | Yes |
| Apple M3 Pro | 4bit | — | 3.0 GB | Yes |
| Apple M1 Pro | 4bit | — | 3.0 GB | Yes |
| Apple M2 Pro | 4bit | — | 3.0 GB | Yes |
| Apple M4 | 4bit | — | 3.0 GB | Yes |
| iMac (Apple M4) | 4bit | — | 3.0 GB | Yes |
| Mac mini (Apple M6) | 4bit | — | 3.0 GB | Yes |
| MacBook Air (Apple M5) | 4bit | — | 3.0 GB | Yes |
| MacBook Pro (Apple M5) | 4bit | — | 3.0 GB | Yes |
| NVIDIA GeForce RTX 5090 | 4bit | 32 GB | 3.0 GB | Yes |
| Apple M2 | 4bit | — | 3.0 GB | Yes |
| Apple M3 | 4bit | — | 3.0 GB | Yes |
| NVIDIA GeForce RTX 3090 | 4bit | 24 GB | 3.0 GB | Yes |
| NVIDIA GeForce RTX 4090 | 4bit | 24 GB | 3.0 GB | Yes |
| Apple M1 | 4bit | — | 3.0 GB | Yes |
Ancestors 1
- gemma-3-4b-ptGoogle
This model
Gemma 3 4B
4.3B params
Quantizations 0
None recorded.
Descendants 0
None recorded.
Papers 24
- arXiv:1905.07830Active35
- arXiv:1905.10044Active35
- arXiv:2104.12756Active35
- arXiv:2502.12404Active35
- arXiv:2203.10244Active35
- arXiv:2502.21228Active35
- arXiv:1910.11856Active35
- arXiv:1904.09728Active35
- arXiv:2103.03874Active35
- arXiv:2106.03193Active35
- arXiv:2404.16816Active35
- arXiv:1810.12440Active35
- arXiv:1911.01547Active35
- arXiv:2312.11805Active35
- arXiv:2311.16502Active35
- arXiv:1908.02660Active35
- arXiv:2107.03374Active35
- arXiv:2009.03300Active35
- arXiv:2210.03057Active35
- arXiv:2108.07732Active35
- arXiv:2311.12022Active35
- arXiv:1705.03551Active35
- arXiv:2110.14168Active35
- arXiv:1911.11641Active35
Repositories 0
No repositories linked yet.
As of
Rewind the record: see this entity's attributes exactly as AI Atlas knew them on a given day.
Claim history · Downloads
Downloadsmetric.downloads1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| 1,682,782 | → current | current | Hugging Face Hub (public pages, model cards, papers)T2 | medium | deterministic |
Claims are temporal and append-only: a new observation closes the previous claim (valid_to) instead of overwriting it. Conflicting claims from different sources are kept side by side and flagged — never averaged. Methodology →
Gemma 3 4B: openness changed from open-weights to restricted
Opennessopen-weights→restrictedhuggingfaceGemma 3 4B scores 5.29% on Humanity's Last Exam
artificial_analysisGemma 3 4B scores 4.83 on Artificial Analysis Intelligence Index
artificial_analysisGemma 3 4B: context length changed from 131072 to 128000
Context window131.1K tokens→128K tokensartificial_analysisGemma 3 4B: openness changed from restricted to open-weights
Opennessrestricted→open-weightsartificial_analysisGemma 3 4B: release date changed from 2025-02-20 to 2025-03-13
Release date20 Feb 2025→13 Mar 2025openrouterGemma 3 4B: release date changed from 2025-03-13 to 2025-02-20
Release date13 Mar 2025→20 Feb 2025huggingfaceGoogle Gemini API lists Gemma 3 4B at $0.05 in / $0.1 out per 1M tokens
openrouter
| Source | Document | Type | Tier | Last observed | Snapshots |
|---|---|---|---|---|---|
| Hugging Face Hub (public pages, model cards, papers) | huggingface.co/google/gemma-3-4b-it | model_page | T2· Quality secondary | 2 min ago | 2 |
| Hugging Face Hub (public pages, model cards, papers) | huggingface.co/models?author=google&p=0&sort=downloads | listing | T2· Quality secondary | 18 min ago | 3 |
| Artificial Analysis | artificialanalysis.ai/leaderboards/models | leaderboard | T2· Quality secondary | 28 min ago | 1 |
| OpenRouter public model & pricing listing | openrouter.ai/api/v1/models | catalogue | T2· Quality secondary | 32 min ago | 2 |
Tier 1 = official/primary, 2 = quality secondary, 3 = community, 4 = unverified. Every snapshot is archived; see all sources and the methodology.