Skip to content
AI Atlas
ModelActiveOpen weightsIdentity probable

Gemma 4 26B A4B

Googlefamily · Gemma 4huggingface.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...

Open in Graph
data quality72

Updated 57 min ago · first seen 11 Sept 2026

model_01M294WW1FFYS3E90RJS36V8HP

Overview

Identity

Canonical model
Yesidentity confidence: mediumOne row per real model release. Artifacts (checkpoints, quantisations, conversions) and folded evaluation variants point here.
Official checkpoints
official_checkpoints = hf_repo identifiers carried by the model itself; artifacts are separate entities pointing here through canonical_id.
Artifacts
3 quantizations0 official · 2 third-partySeparate entities (checkpoint · quantization · conversion · packaging) pointing to this model through canonical_id.
Provider deployments
2
API aliases
gemma-4-26b-a4bgemma-4-26b-a4b-non-reasoninggoogle/gemma-4-26b-a4b-itIdentifiers under which providers and evaluators refer to this model.
Folded evaluation variants
2Effort / thinking variants (…-high, …-non-reasoning) are result configurations of this model, not separate models. Their old URLs redirect here.

Openness

Open weights— weights downloadable under Apache-2.0; commercial use allowed; redistribution allowed; derivatives allowed; 4 dimensions unknown.

Weights downloadable under a permissive or Creative Commons licence allowing commercial use; code or data may be missing.

  • Weights

    Yes

  • Inference code

    —

  • Training code

    —

  • Training data

    —

  • Dataset

    —

  • Commercial use

    Yes

  • Redistribution

    Yes

  • Derivatives

    Yes

Licence: Apache License 2.0 (permissive · SPDX Apache-2.0 · stated as “apache-2.0”)

dimensions marked null are unknown, not false

Key facts

Release date

Source:Hugging Face Hub (public pages, model cards, papers)T2observed 12 h agomedium

Openrouter id

Source:OpenRouter public model & pricing listingT2observed 16 h agomedium

Architecture

Architecture

Source:Hugging Face Hub (public pages, model cards, papers)T2observed 16 h agomedium

Model type

Source:Hugging Face Hub (public pages, model cards, papers)T2observed 16 h agomedium

Parameters

Source:Hugging Face Hub (public pages, model cards, papers)T2observed 16 h agomedium

Active parameters

Source:Hugging Face Hub (public pages, model cards, papers)T2observed 16 h agomedium

Tokenizer

Source:OpenRouter public model & pricing listingT2observed 16 h agomedium

Weights dtype

Source:Hugging Face Hub (public pages, model cards, papers)T2observed 16 h agomedium

File size

Source:Hugging Face Hub (public pages, model cards, papers)T2observed 16 h agomedium

Library name

Source:Hugging Face Hub (public pages, model cards, papers)T2observed 16 h agomedium

Pipeline tag

Source:Hugging Face Hub (public pages, model cards, papers)T2observed 16 h agomedium

Hugging Face repo

Source:Hugging Face Hub (public pages, model cards, papers)T2observed 2 h agomedium

Capabilities

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

    Artificial Analysis · T2

  • Vision

    Yes

    OpenRouter public model & pricing listing · T2

  • Audio

    Unavailable

  • Fine-tuning available

    Unavailable

Context window

Source:OpenRouter public model & pricing listingT2observed 13 h agomedium

Max output

Source:OpenRouter public model & pricing listingT2observed 16 h agomedium

Tokenizer

Source:OpenRouter public model & pricing listingT2observed 16 h agomedium

Benchmarks30

Compare with another model →

Comparable same task and conditions · Partially comparable same task, conditions differ (effort, temperature, judge) · Not comparable different variant or metric

Benchmark results grouped by comparability group
Benchmark · groupBest scoreTrustConfigurationResultsvs leaderEvaluatedSource
Terminal-Benchagentic · accuracy · variant=hard · evaluator=Artificial AnalysisIndependentevaluatorArtificial Analysisvarianthardreasoningoffconditions differ across rows → partially comparable4−40.9 ptvs gpt-5.6-solobs. 12 Sept 2026artificialanalysis.aiT2
Terminal-Benchagentic · accuracy · variant=v2.1 · evaluator=Artificial AnalysisIndependentevaluatorArtificial Analysisvariantv2.1reasoningonconditions differ across rows → partially comparable2non-primary groupobs. 12 Sept 2026artificialanalysis.aiT2
τ²-benchagentic · pass^1 · variant=Telecom · evaluator=Artificial AnalysisIndependentevaluatorArtificial AnalysisvariantTelecomreasoningonconditions differ across rows → partially comparable2non-primary groupobs. 12 Sept 2026artificialanalysis.aiT2
τ²-benchagentic · pass^1 · evaluator=Artificial AnalysisIndependentevaluatorArtificial Analysis2−55.6 ptvs Z.ai GLM 5.2obs. 11 Sept 2026artificialanalysis.aiT2
Artificial Analysis Intelligence Indexcomposite · indexIndependentreasoningonversion4.3conditions differ across rows → partially comparable4−36.7vs Claude Fable 5.1obs. 12 Sept 2026artificialanalysis.aiT2
IFBenchinstruction-following · accuracy · evaluator=Artificial AnalysisIndependentevaluatorArtificial Analysisreasoningonconditions differ across rows → partially comparable4−10.9 ptvs Grok 4.3obs. 12 Sept 2026artificialanalysis.aiT2
Humanity's Last Examknowledge · accuracy · evaluator=Artificial AnalysisIndependentevaluatorArtificial Analysisreasoningonconditions differ across rows → partially comparable4−39.8 ptvs Claude Fable 5.1obs. 12 Sept 2026artificialanalysis.aiT2
MMMU-Promultimodal · accuracy · evaluator=Artificial AnalysisIndependentevaluatorArtificial Analysisreasoningonconditions differ across rows → partially comparable4−17.6 ptvs gpt-6-astraobs. 12 Sept 2026artificialanalysis.aiT2
GPQA Diamondreasoning · accuracy · variant=Diamond · evaluator=Artificial AnalysisIndependentevaluatorArtificial AnalysisvariantDiamondreasoningonconditions differ across rows → partially comparable2non-primary groupobs. 12 Sept 2026artificialanalysis.aiT2
GPQA Diamondreasoning · accuracy · variant=GPQA Diamond · evaluator=Artificial AnalysisIndependentevaluatorArtificial AnalysisvariantGPQA Diamond2−17.1 ptvs gpt-6-astraobs. 11 Sept 2026artificialanalysis.aiT2

Current rows only, grouped by benchmark → canonical metric → comparability group (task configuration). Effort variants folded into this model appear as rows of the same group. 30 current rows in total. “vs leader” compares with the current leader of the benchmark's primary group only; other groups are not directly comparable. Comparability rules →

Providers & Pricing4

All offers in the price terminal →
Provider deployments, cheapest output first
ProviderContextInput / 1MOutput / 1MStatusObservedSource
Google Gemini APIcheapest outputgoogle/gemma-4-26b-a4b-it:free262.1Kout 32.8Kactive7 h agosince 11 Sept 2026openrouter.aiT2
OpenRoutergoogle/gemma-4-26b-a4b-it:free262.1Kout 32.8Kactive57 min agosince 12 Sept 2026openrouter.aiT2
Google Gemini APIgoogle/gemma-4-26b-a4b-it131.1Kout 32.8Kactive7 h agosince 11 Sept 2026openrouter.aiT2
OpenRoutergoogle/gemma-4-26b-a4b-it131.1Kout 32.8Kactive57 min agosince 12 Sept 2026openrouter.aiT2

USD per 1M tokens as published by each provider; native units (per-request fees, flex/priority tiers) are kept verbatim. Rows are append-only — every price change is kept in the history below. Cost of a workload →

Price history

Step lines per provider; amber markers are recorded changes. Click a marker or a row for the evidence behind that price.

Output price · USD / 1M tokens 2 providers

Output price history of Gemma 4 26B A4B$0$0.10$0.20$0.30Sept 26Sept 26Sept 26Sept 26Sept 26Google Gemini API: first observed → $0.22 · 11 Sept 2026Google Gemini API: $0.22 → $0 · 11 Sept 2026OpenRouter: first observed → $0.22 · 12 Sept 2026OpenRouter: $0.22 → $0 · 12 Sept 2026
  • Google Gemini API
  • OpenRouter
  • OpenRouter$0.22 → $012 Sept 2026
  • OpenRouterfirst observed $0.2212 Sept 2026
  • Google Gemini API$0.22 → $011 Sept 2026
  • Google Gemini APIfirst observed $0.2211 Sept 2026

Input price · USD / 1M tokens 2 providers

Input price history of Gemma 4 26B A4B$0$0.01$0.02$0.03$0.04$0.05Sept 26Sept 26Sept 26Sept 26Sept 26Google Gemini API: first observed → $0.042 · 11 Sept 2026Google Gemini API: $0.042 → $0 · 11 Sept 2026OpenRouter: first observed → $0.042 · 12 Sept 2026OpenRouter: $0.042 → $0 · 12 Sept 2026
  • Google Gemini API
  • OpenRouter
  • OpenRouter$0.042 → $012 Sept 2026
  • OpenRouterfirst observed $0.04212 Sept 2026
  • Google Gemini API$0.042 → $011 Sept 2026
  • Google Gemini APIfirst observed $0.04211 Sept 2026

Hardware fit37

Estimated

36 of 37 device × quantization combinations fit.

Run locally: your machine →
Estimated hardware fit
HardwareQuantizationDevice memoryEst. memoryFits
Apple M3 Ultra4bit—15.3 GB est.Yes
Mac Studio (Apple M5 Ultra)4bit—15.3 GB est.Yes
Apple M2 Ultra4bit—15.3 GB est.Yes
Apple M1 Ultra4bit—15.3 GB est.Yes
Apple M3 Max4bit—15.3 GB est.Yes
Apple M4 Max4bit—15.3 GB est.Yes
Mac Studio (Apple M5 Max)4bit—15.3 GB est.Yes
MacBook Pro (Apple M5 Max)4bit—15.3 GB est.Yes
Apple M2 Max4bit—15.3 GB est.Yes
Apple M1 Max4bit—15.3 GB est.Yes
Apple M4 Pro4bit—15.3 GB est.Yes
Mac mini (Apple M5 Pro)4bit—15.3 GB est.Yes
MacBook Pro (Apple M5 Pro)4bit—15.3 GB est.Yes
Apple M3 Pro4bit—15.3 GB est.Yes
Apple M1 Pro4bit—15.3 GB est.Yes
Apple M2 Pro4bit—15.3 GB est.Yes
Apple M44bit—15.3 GB est.Yes
iMac (Apple M4)4bit—15.3 GB est.Yes
Mac mini (Apple M6)4bit—15.3 GB est.Yes
MacBook Air (Apple M5)4bit—15.3 GB est.Yes
MacBook Pro (Apple M5)4bit—15.3 GB est.Yes
Apple M24bit—15.3 GB est.Yes
Apple M34bit—15.3 GB est.Yes
NVIDIA GeForce RTX 30904bit24 GB15.3 GB est.Yes
NVIDIA GeForce RTX 40904bit24 GB15.3 GB est.Yes
NVIDIA GeForce RTX 50904bit32 GB15.3 GB est.Yes
NVIDIA A100 80GB4bit80 GB15.3 GB est.Yes
NVIDIA H100 SXM4bit80 GB15.3 GB est.Yes
NVIDIA H100 NVL4bit94 GB15.3 GB est.Yes
NVIDIA DGX Spark4bit128 GB15.3 GB est.Yes
NVIDIA H2004bit141 GB15.3 GB est.Yes
NVIDIA H200 NVL4bit141 GB15.3 GB est.Yes
NVIDIA B2004bit180 GB15.3 GB est.Yes
AMD Instinct MI300X4bit192 GB15.3 GB est.Yes
AMD Instinct MI325X4bit256 GB15.3 GB est.Yes
NVIDIA DGX B2004bit1,440 GB15.3 GB est.Yes
Apple M14bit—15.3 GB est.No
Assumptions (7)
  • Estimated, not measured: weights = parameters × bytes/param × 1.15 runtime overhead (or the observed artifact file size when one is recorded).
  • bytes/param: 4bit = 0.5, 8bit = 1.0, fp16 = 2.0 (uniform quantization, no per-layer exceptions).
  • KV cache: 2 × layers × kv_heads × head_dim × 2 bytes × context × batch when the architecture is known; otherwise 0.5 GB per 8 192 tokens (× batch), independent of architecture (GQA/MLA models need less).
  • A model 'fits' when the estimate is at most the device memory minus 2 GB reserved for the OS and framework.
  • Mixture-of-experts models are estimated on total parameters (all experts must be resident); active parameters are ignored.
  • Device memory uses the largest configuration when several are listed (e.g. Apple silicon tiers).
  • Multi-GPU: device memories are summed; interconnect bandwidth, tensor-parallel replication and pipeline bubbles are not modelled.

Lineage

Open in Graph →
Explicit derived_from / fine_tuned_from / distilled_from relations stated by sources; artifacts collapsed by kind.
DESCENDANTS 0 · ARTIFACTS3 quantizationsartifacts · collapsed3 quantizations — artifacts · collapsedGemma 4 26B A4B25.8B params · this modelGemma 4 26B A4B — 25.8B params · this model

    Versions & Artifacts3

    Version history

    Context window2 changes

    11 Sept 2026→11 Sept 2026→12 Sept 2026current

    Licensefirst observation only

    11 Sept 2026current

    Max outputfirst observation only

    11 Sept 2026current

    Opennessfirst observation only

    11 Sept 2026current

    Parametersfirst observation only

    11 Sept 2026current

    Each hop is a claim: click a value for its source, tier and observation time. Nothing is overwritten — a new observation closes the previous claim.

    Artifacts 3

    quantizations 3

    Papers1

    Timeline27

    Full timeline →

    Change history

    Temporal, append-only claims: a new observation closes the previous claim instead of overwriting it. Rewind the record with the as-of picker.
    1 claims · 1 propertiesShow all properties

    Accessaccess1

    Claim history for Access
    ValueValid from → toStatusSourceConfidenceExtractor
    open → currentcurrentHugging Face Hub (public pages, model cards, papers)T2mediumdeterministic

    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 →

    Provenance

    Attributed facts

    47

    Source tiers

    T247

    Freshest observation

    57 min ago

    Conflicts

    None

    Source documents 7

    Source documents
    SourceDocumentTypeTierLast observedSnapshots
    OpenRouter public model & pricing listingopenrouter.ai/api/v1/models catalogueT2· Quality secondary57 min ago12
    Hugging Face Hub (public pages, model cards, papers)huggingface.co/google/gemma-4-26B-A4B-it/raw/main/README.md model_cardT2· Quality secondary1 h ago1
    Hugging Face Hub (public pages, model cards, papers)huggingface.co/unsloth/gemma-4-26B-A4B-it-GGUF model_pageT2· Quality secondary2 h ago5
    Hugging Face Hub (public pages, model cards, papers)huggingface.co/nvidia/Gemma-4-26B-A4B-NVFP4 model_pageT2· Quality secondary2 h ago5
    Hugging Face Hub (public pages, model cards, papers)huggingface.co/google/gemma-4-26B-A4B-it model_pageT2· Quality secondary2 h ago5
    Hugging Face Hub (public pages, model cards, papers)huggingface.co/models?author=google&p=0&sort=downloads listingT2· Quality secondary2 h ago8
    Artificial Analysisartificialanalysis.ai/leaderboards/models leaderboardT2· Quality secondary6 h ago2

    Tier 1 = official/primary, 2 = quality secondary, 3 = community, 4 = unverified. Every snapshot is archived; see all sources and the methodology.

    Data quality (72/100) measures how well AI Atlas knows this entity — completeness, primary-source ratio, freshness, conflicts — never how good the model is. Methodology →