Gemma 4 26B A4B
Googlehuggingface.co/google/gemma-4-26B-A4B-it
Gemma 4 26B A4B IT is an instruction-tuned Mixture-of-Experts (MoE) model from Google DeepMind. Despite 25.2B total parameters, only 3.8B activate per token during inference — delivering near-31B quality at...
Updated 2 h ago · first seen 11 Sept 2026
model_01M294WW1FFYS3E90RJS36V8HP
- Parameters
- 25.8B · 4B active
- T2 · 5 h ago
- Context
- 262.1K tokens
- T2 · 2 h ago
- Released
- 3 Apr 2026
- T2 · 2 h ago
- License
- apache-2.0
- T2 · 4 h ago
Specification
- Release date
- 3 Apr 2026
Source:OpenRouter public model & pricing listingT2observed 2 h agomedium
- Openness
- open-weights
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 5 h agomedium
- License
- apache-2.0
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 4 h agomedium
- Architecture
- Gemma4ForConditionalGeneration
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 4 h agomedium
- Parameters
- 25.8B
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 5 h agomedium
- Active parameters
- 4B
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 4 h agomedium
- Context window
- 262.1K tokens
Source:OpenRouter public model & pricing listingT2observed 2 h agomedium
- Max output
- 32.8K tokens
Source:OpenRouter public model & pricing listingT2observed 5 h agomedium
- Modalities
- image, text, video
Source:OpenRouter public model & pricing listingT2observed 5 h agomedium
- Input modalities
- image, text, video
Source:OpenRouter public model & pricing listingT2observed 5 h agomedium
- Output modalities
- text
Source:OpenRouter public model & pricing listingT2observed 5 h agomedium
- Tokenizer
- Gemma
Source:OpenRouter public model & pricing listingT2observed 5 h agomedium
- Base model
- google/gemma-4-26B-A4B
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 4 h agomedium
- File size
- 51.6 GB
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 4 h agomedium
- Hugging Face repo
- google/gemma-4-26B-A4B-it
Source:OpenRouter public model & pricing listingT2observed 5 h agomedium
- Pipeline tag
- image-text-to-text
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 5 h agomedium
- Model card
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 5 h agomedium
- Downloads
- 8,983,063
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 3 h agomedium
- Likes
- 1,489
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 3 h agomedium
- Gated
- No
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 4 h agomedium
- Hf inference providers
- deepinfra, featherless-ai, novita, scaleway
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 4 h agomedium
- Last modified
- 2026-07-20T16:42:12+00:00
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 5 h agomedium
- Library name
- transformers
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 4 h agomedium
- License url
- https://ai.google.dev/gemma/docs/gemma_4_license
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 4 h agomedium
- Downloads all time
- 58,013,646
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 3 h agomedium
- Model type
- gemma4
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 4 h agomedium
- Openrouter id
- google/gemma-4-26b-a4b-it
Source:OpenRouter public model & pricing listingT2observed 5 h agomedium
- Supported parameters
- frequency_penalty, include_reasoning, logit_bias, logprobs, max_tokens, min_p, presence_penalty, reasoning, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k, top_logprobs, top_p
Source:OpenRouter public model & pricing listingT2observed 5 h agomedium
- Tags
- transformers, safetensors, gemma4, image-text-to-text
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 4 h agomedium
- Weights dtype
- BF16
Source:Hugging Face Hub (public pages, model cards, papers)T2observed 4 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
35
Source tiers
T235
Freshest observation
2 h ago
Conflicts
None
Modalities
- Modalities
- imagetextvideo
- Input
- imagetextvideo
- Output
- text
Capabilities
Tool calling
Yes
OpenRouter public model & pricing listing · T2
Structured output
Yes
OpenRouter public model & pricing listing · T2
Reasoning
Yes
OpenRouter public model & pricing listing · T2
Vision
Yes
OpenRouter public model & pricing listing · T2
Audio
Unavailable
Fine-tuning available
Unavailable
- Context window
- 262.1K tokens
Source:OpenRouter public model & pricing listingT2observed 2 h agomedium
- Max output
- 32.8K tokens
Source:OpenRouter public model & pricing listingT2observed 5 h agomedium
- Tokenizer
- Gemma
Source:OpenRouter public model & pricing listingT2observed 5 h agomedium
| Benchmark | Score | Metric | Config | Evaluated | Source |
|---|---|---|---|---|---|
| τ²-bench | 43.57% | pass^1 | aa_slug=gemma-4-26b-a4b · evaluator=Artificial Analysis · index_version=4.3 | — | artificialanalysis.aiT2 |
| Artificial Analysis Intelligence Index | 16.67 | index | aa_slug=gemma-4-26b-a4b · version=4.3 · estimated=true | — | artificialanalysis.aiT2 |
| GPQA | 79.19% | accuracy | aa_slug=gemma-4-26b-a4b · variant=GPQA Diamond · evaluator=Artificial Analysis · index_version=4.3 | — | artificialanalysis.aiT2 |
| Humanity's Last Exam | 19.32% | accuracy | aa_slug=gemma-4-26b-a4b · evaluator=Artificial Analysis · index_version=4.3 | — | artificialanalysis.aiT2 |
| IFBench | 72.45% | accuracy | aa_slug=gemma-4-26b-a4b · evaluator=Artificial Analysis · index_version=4.3 | — | artificialanalysis.aiT2 |
| MMMU-Pro | 69.25% | accuracy | aa_slug=gemma-4-26b-a4b · evaluator=Artificial Analysis · index_version=4.3 | — | artificialanalysis.aiT2 |
| Terminal-Bench | 38.95% | accuracy | aa_slug=gemma-4-26b-a4b · variant=v2.1 · evaluator=Artificial Analysis · index_version=4.3 | — | artificialanalysis.aiT2 |
| Terminal-Bench | 13.64% | accuracy | aa_slug=gemma-4-26b-a4b · 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-4-26b-a4b-it:free | $0 | $0 | — | — | 262.1K | 2 h ago | openrouter.aiT2 |
| Google Gemini APIgoogle/gemma-4-26b-a4b-it | $0.042 | $0.22 | — | — | 131.1K | 2 h 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 | 15.3 GB | Yes |
| Apple M3 Ultra | 4bit | — | 15.3 GB | Yes |
| Mac Studio (Apple M5 Ultra) | 4bit | — | 15.3 GB | Yes |
| AMD Instinct MI325X | 4bit | 256 GB | 15.3 GB | Yes |
| AMD Instinct MI300X | 4bit | 192 GB | 15.3 GB | Yes |
| Apple M2 Ultra | 4bit | — | 15.3 GB | Yes |
| NVIDIA B200 | 4bit | 180 GB | 15.3 GB | Yes |
| NVIDIA H200 | 4bit | 141 GB | 15.3 GB | Yes |
| NVIDIA H200 NVL | 4bit | 141 GB | 15.3 GB | Yes |
| Apple M1 Ultra | 4bit | — | 15.3 GB | Yes |
| Apple M3 Max | 4bit | — | 15.3 GB | Yes |
| Apple M4 Max | 4bit | — | 15.3 GB | Yes |
| Mac Studio (Apple M5 Max) | 4bit | — | 15.3 GB | Yes |
| MacBook Pro (Apple M5 Max) | 4bit | — | 15.3 GB | Yes |
| NVIDIA DGX Spark | 4bit | 128 GB | 15.3 GB | Yes |
| Apple M2 Max | 4bit | — | 15.3 GB | Yes |
| NVIDIA H100 NVL | 4bit | 94 GB | 15.3 GB | Yes |
| NVIDIA A100 80GB | 4bit | 80 GB | 15.3 GB | Yes |
| NVIDIA H100 SXM | 4bit | 80 GB | 15.3 GB | Yes |
| Apple M1 Max | 4bit | — | 15.3 GB | Yes |
| Apple M4 Pro | 4bit | — | 15.3 GB | Yes |
| Mac mini (Apple M5 Pro) | 4bit | — | 15.3 GB | Yes |
| MacBook Pro (Apple M5 Pro) | 4bit | — | 15.3 GB | Yes |
| Apple M3 Pro | 4bit | — | 15.3 GB | Yes |
| Apple M1 Pro | 4bit | — | 15.3 GB | Yes |
| Apple M2 Pro | 4bit | — | 15.3 GB | Yes |
| Apple M4 | 4bit | — | 15.3 GB | Yes |
| iMac (Apple M4) | 4bit | — | 15.3 GB | Yes |
| Mac mini (Apple M6) | 4bit | — | 15.3 GB | Yes |
| MacBook Air (Apple M5) | 4bit | — | 15.3 GB | Yes |
| MacBook Pro (Apple M5) | 4bit | — | 15.3 GB | Yes |
| NVIDIA GeForce RTX 5090 | 4bit | 32 GB | 15.3 GB | Yes |
| Apple M2 | 4bit | — | 15.3 GB | Yes |
| Apple M3 | 4bit | — | 15.3 GB | Yes |
| NVIDIA GeForce RTX 3090 | 4bit | 24 GB | 15.3 GB | Yes |
| NVIDIA GeForce RTX 4090 | 4bit | 24 GB | 15.3 GB | Yes |
| Apple M1 | 4bit | — | 15.3 GB | No |
Ancestors 1
- gemma-4-26B-A4BGoogle
This model
Gemma 4 26B A4B
25.8B params
Quantizations 2
Descendants 0
None recorded.
Papers 1
- arXiv:2607.02770Active35
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 · License url
License urllicense_url1
| Value | Valid from → to | Status | Source | Confidence | Extractor |
|---|---|---|---|---|---|
| https://ai.google.dev/gemma/docs/gemma_4_license | → 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 4 26B A4B: context length changed from 256000 to 262144
Context window256K tokens→262.1K tokensopenrouterGemma 4 26B A4B scores 13.64% on Terminal-Bench
artificial_analysisGemma 4 26B A4B scores 38.95% on Terminal-Bench
artificial_analysisGemma 4 26B A4B scores 43.57% on τ²-bench
artificial_analysisGemma 4 26B A4B scores 69.25% on MMMU-Pro
artificial_analysisGemma 4 26B A4B scores 72.45% on IFBench
artificial_analysisGemma 4 26B A4B scores 19.32% on Humanity's Last Exam
artificial_analysisGemma 4 26B A4B scores 16.67 on Artificial Analysis Intelligence Index
artificial_analysisGemma 4 26B A4B: context length changed from 262144 to 256000
Context window262.1K tokens→256K tokensartificial_analysisGemma 4 26B A4B: release date changed from 2026-03-11 to 2026-04-03
Release date11 Mar 2026→3 Apr 2026openrouterGemma 4 26B A4B: release date changed from 2026-04-03 to 2026-03-11
Release date3 Apr 2026→11 Mar 2026huggingfaceGoogle Gemini API lists Gemma 4 26B A4B at $0 in / $0 out per 1M tokens
openrouterGoogle Gemini API lists Gemma 4 26B A4B at $0.042 in / $0.22 out per 1M tokens
openrouter
| Source | Document | Type | Tier | Last observed | Snapshots |
|---|---|---|---|---|---|
| OpenRouter public model & pricing listing | openrouter.ai/api/v1/models | catalogue | T2· Quality secondary | 2 h ago | 3 |
| Hugging Face Hub (public pages, model cards, papers) | huggingface.co/google/gemma-4-26B-A4B-it/raw/main/README.md | model_card | T2· Quality secondary | 2 h ago | 1 |
| Hugging Face Hub (public pages, model cards, papers) | huggingface.co/unsloth/gemma-4-26B-A4B-it-GGUF | model_page | T2· Quality secondary | 2 h ago | 2 |
| Hugging Face Hub (public pages, model cards, papers) | huggingface.co/nvidia/Gemma-4-26B-A4B-NVFP4 | model_page | T2· Quality secondary | 2 h ago | 2 |
| Hugging Face Hub (public pages, model cards, papers) | huggingface.co/google/gemma-4-26B-A4B-it | model_page | T2· Quality secondary | 3 h ago | 2 |
| Hugging Face Hub (public pages, model cards, papers) | huggingface.co/models?author=google&p=0&sort=downloads | listing | T2· Quality secondary | 3 h ago | 3 |
| Artificial Analysis | artificialanalysis.ai/leaderboards/models | leaderboard | T2· Quality secondary | 3 h ago | 1 |
Tier 1 = official/primary, 2 = quality secondary, 3 = community, 4 = unverified. Every snapshot is archived; see all sources and the methodology.