Google DeepMindopen-weight
Released: 2026-04-02Artificial AnalysisGemma 4 26B A4B
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...
Intelligence Index
—/ 100
Artificial Analysis Intelligence Index (independent composite)Throughput Speed
71tok/sec
Output streaming throughputTime to First Answer
26s
Time to first answer token, including any thinkingCost to Run the Index
—USD
Artificial Analysis cost of the full Index runMeasured results · Artificial Analysis
Independent evaluations of Gemma 4 26B A4B (Reasoning configuration). Values are reported as published; nothing here is estimated or derived. Snapshot 2026-11-15.
- Intelligence Index
- —
- Cost to run the Index
- —
- Time per task
- —
- Omniscience index
- —
- Reasoning model
- Yes
- Cached-input price
- $0.05 / 1M
Reasoning & knowledge
- CritPt0.0%
- GPQA Diamond79.2%
- HLE-Diamond6.2%
- Humanity's Last Exam19.3%
- IFBench72.5%
Coding & terminal agents
- SciCode40.1%
- Terminal-Bench 2.139.0%
- Terminal-Bench Hard13.6%
Agentic work
- AA-Briefcase v1.1 (Elo)316 Elo
- AutomationBench-AA1.9%
- GDPval-AA (normalized Elo)573 Elo
- ITBench-AA23.6%
- τ²-Bench Telecom43.6%
- τ³-Banking12.0%
Long context & documents
Multimodal & science
- MMMU-Pro69.3%
Reasoning-effort variants evaluated
| Variant | Intelligence Index | Cost to run Index | Output speed (tok/s) |
|---|---|---|---|
| Reasoning | — | — | 71.4 |
| Non-reasoning | — | — | 63.0 |
Sourced Benchmark Evaluations (14)
Verified performance across authoritative benchmarks with provenance tracking.| Benchmark | Status | Score | Trust | Date | Source Type | Provenance |
|---|---|---|---|---|---|---|
| GPQA | Nearing Saturation | 79.2% | 78 | 2026-11-15 | independent | Source ↗ |
| Humanity's Last Exam | Active | 19.3% | 93 | 2026-11-15 | independent | Source ↗ |
| MMMU-Pro | Active | 69.2% | 93 | 2026-11-15 | independent | Source ↗ |
| IFBench | Active | 72.4% | 93 | 2026-11-15 | independent | Source ↗ |
| SciCode | Active | 40% | 93 | 2026-11-15 | independent | Source ↗ |
| τ²-Bench Telecom | Active | 43.6% | 93 | 2026-11-15 | independent | Source ↗ |
| τ³-Banking | Active | 12% | 93 | 2026-11-15 | independent | Source ↗ |
| Terminal-Bench Hard | Active | 13.6% | 93 | 2026-11-15 | independent | Source ↗ |
| Terminal-Bench 2.1 | Active | 39% | 93 | 2026-11-15 | independent | Source ↗ |
| AA-LCR (Long Context Reasoning) | Active | 65.7% | 93 | 2026-11-15 | independent | Source ↗ |
| CritPt | Active | 0% | 93 | 2026-11-15 | independent | Source ↗ |
| HLE-Diamond | Active | 6.2% | 93 | 2026-11-15 | independent | Source ↗ |
| AutomationBench-AA | Active | 1.9% | 93 | 2026-11-15 | independent | Source ↗ |
| ITBench-AA | Active | 23.6% | 93 | 2026-11-15 | independent | Source ↗ |
Local Hardware Execution (Ollama)Hardware Compatible
# 1. Pull and execute model locally with Ollama
ollama run google-gemma-4-26b-a4b
# 2. Or invoke via local OpenAI-compatible endpoint
curl http://localhost:11434/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "google-gemma-4-26b-a4b",
"messages": [{"role": "user", "content": "Analyze reasoning chains on AIME 2026"}]
}'Evaluation & Sourcing Notes
Scores listed for Gemma 4 26B A4B represent verified evaluations extracted from official research papers, independent evaluation suites (HELM, LMSYS, OpenCompass, LiveBench), and verified audit reports.
All benchmarks marked as saturated or deprecated reflect historical performance where the benchmark no longer provides active discriminative power.