HOST DIAGNOSTIC · Google
Gemma 4 E2B
Gemma 4 E2B· Gemma· OPEN WEIGHTS
ACTIVE
HOST IDENTIFICATION
Google
Gemma 4 E2B
ACTIVE
AVG SCORE
[12.1]
CAPABLE
STRONGEST
Cost Efficiency
[20]
WEAKEST
Code Generation
[10]
MAXED (20/20)
1
of 20 attributes
20-DIMENSION EVALUATION
ATTRIBUTE SCORES · 20 DIMENSIONS
Cognitive
Bulk Apperception
11
Reasoning
11
Mathematical Precision
11
World Knowledge
11
Scientific Acumen
11
Technical
Code Generation
10
Tool Use
10
Multimodal Fluency
12
Speed
18
Cost Efficiency
20
Behavioral
Candor
13
Creativity
11
Tenacity
10
Self-Correction
10
Calibration
13
Operational
Instruction Following
11
Context Fidelity
13
Multi-turn Coherence
11
Planning and Execution
10
Safety Alignment
15
SCORING METHODOLOGY
CALIBRATED. Google DeepMind, released 2026-04-02, Apache 2.0 open-weights. Effective-2B (E2B) smallest variant — phone/IoT/very-low-latency tier; runs in ~4GB RAM. 128K context, native audio + image input, configurable thinking, MTP. The 'democratizing access' end of the family. DIMENSION RATIONALE: Cost Efficiency 20 + Speed 18 max-out (pocket-deployable, near-zero marginal cost); capability dims 10-12 reflect the smallest-model raw-intelligence floor while still being a capable reasoner for size (Google's 'intelligence-per-parameter' framing). Multimodal 12. Confidence: LOW — scored as the family floor by scaling from 31B anchor; no dedicated E2B leaderboard run located. Flag for refresh when on-device benchmarks publish (e.g. Android Bench LLM).