HOST DIAGNOSTIC · Google
Gemma 4 26B A4B
Gemma 4 26B A4B· Gemma· OPEN WEIGHTS
ACTIVE
HOST IDENTIFICATION
Google
Gemma 4 26B A4B
ACTIVE
AVG SCORE
[14.0]
STRONG
STRONGEST
Cost Efficiency
[19]
WEAKEST
Code Generation
[13]
MAXED (20/20)
0
of 20 attributes
20-DIMENSION EVALUATION
ATTRIBUTE SCORES · 20 DIMENSIONS
Cognitive
Bulk Apperception
14
Reasoning
14
Mathematical Precision
14
World Knowledge
13
Scientific Acumen
14
Technical
Code Generation
13
Tool Use
13
Multimodal Fluency
13
Speed
16
Cost Efficiency
19
Behavioral
Candor
14
Creativity
13
Tenacity
13
Self-Correction
13
Calibration
14
Operational
Instruction Following
14
Context Fidelity
14
Multi-turn Coherence
14
Planning and Execution
13
Safety Alignment
15
SCORING METHODOLOGY
CALIBRATED. Google DeepMind, released 2026-04-02, Apache 2.0 open-weights. 26B-total / ~3.8-4B-active Mixture-of-Experts (MoE) variant; Google's stated DEFAULT recommendation for many tasks ('start with 26B A4B'). Ranks #6 open model on Arena AI text (parent family note). 256K context, multimodal, configurable thinking, MTP speculative decoding. Runs on consumer GPUs. DIMENSION RATIONALE: MoE active-param efficiency gives Cost Efficiency 19 + Speed 16 (only ~4B active params → fast/cheap relative to capability). Capability dims 13-14, below the 31B Dense flagship on raw reasoning/code/math but strong intelligence-per-active-param. Multimodal 13 (image in). Anchored between E4B and 31B. Confidence: MEDIUM — Arena #6 placement is provenance; per-benchmark sub-scores for this specific size less published than 31B (scored conservatively where uncertain).