Gemma 4 26B A4B (IT): benchmark results
Instruction-tuned Gemma 4 26B A4B checkpoint. Provider: Google. Released 2026-04-02. Access: Open.
Aggregate Rating 1657 ± 6, rank #297 of 1656 rated models, from 1203 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| KhatianDoc - Field Extraction Row F1 | 25.84 | Row-level F1 (%; structured field extraction over 261 rows i | 100 |
| KrishokChat - Treatment QA (Oracle Context) | 54.05 | Treatment QA correct rate (%): the recommended treatment mat | 100 |
| Liquid LFM2.5 24B Card - HLE | 8.57 | Score (%) | 100 |
| PHANTOM - FC ATTACK ASR | 18.91 | Reported aggregate attack success rate (%) for FC ATTACK att | 100 |
| QIMMA - MedArabiQ QA | 56.88 | BERTScore F1 (%) | 100 |
| SEA-HELM (English) | 72.37 | Normalized Score | 100 |
| SEA-HELM (Thai) - Causal Reasoning | 98.06 | Normalized Score | 100 |
| Ukrainian LLM - ARC Challenge UK | 92.49 | Score (%) | 100 |
| Ukrainian LLM - Belebele UKR Cyrl | 95.33 | Score (%) | 100 |
| Ukrainian LLM - Flores UK | 36.68 | Score (%) | 100 |
| Ukrainian LLM - GSM8K UK | 94.24 | Score (%) | 100 |
| Ukrainian LLM - Global MMLU Full UK | 86.64 | Score (%) | 100 |
Interactive version: theaggregate.ai/model?slug=gemma-4-26b-a4b-it · How It Works · Data refreshed daily, snapshot 2026-10-20.