Automated page. Written and translated automatically by AI and checked against the sources by software; no human editor. SI-IQ is an index built from public benchmark results, not an IQ test. How it works

How smart is Gemma 3 4B?

63 SI-IQ · 90% interval 53–73
Not ranked: too little data

Gemma 3 4B is a model from Google DeepMind, released on Mar 12, 2025. Its SI-IQ is 63 (90% interval 53–73), estimated from 3 benchmark results by 3 independent evaluators.

This model has fewer than 5 independent benchmark results or fewer than 2 different evaluators, so it is shown but not ranked.

Facts

Developer: Google DeepMind
Released: Mar 12, 2025
Access: Open weights (restricted licence)
Benchmarks used: 3 results from 3 evaluators

Ability by area (SI-IQ, 90% interval)

Every benchmark result we used

Each score is one value exactly as the source published it: for Epoch AI the best of its run settings, for OpenRouter the median hosting provider (the developer's own hosting when it ran the test). “Implied SI-IQ” is the ability that this single result points to; the overall SI-IQ is their precision-weighted average.

Benchmark	Area	Score	Implied SI-IQ	Evaluated by	As listed by the source	Retrieved	Source
OTIS Mock AIME 2024–2025	Math	7.5%	67	Epoch AI	`gemma-3-4b-it`	Oct 18, 2026	Epoch AI Benchmarking Hub · data file
GPQA Diamond	Knowledge	23.2%	−33	Epoch AI	`gemma-3-4b-it`	Oct 18, 2026	Epoch AI Benchmarking Hub · data file
GPQA Diamond (OpenRouter run)	Knowledge	27.1%	11	OpenRouter	`google/gemma-3-4b-it` Parasail · median over providers (1 of 1 rows)	Oct 18, 2026	OpenRouter benchmark datasets · data file
Berkeley Function Calling Leaderboard	Agents	19.6%	53	UC Berkeley (Gorilla)	`Gemma-3-4b-it (Prompt)`	Oct 18, 2026	Berkeley Function Calling Leaderboard · data file

Human preference (Arena)

On the Arena leaderboard, where people vote between two anonymous answers, Gemma 3 4B has a rating of 1,303 (1,294–1,313) from 4,171 votes (published Oct 15, 2026). Human preference is a different measure from benchmark ability and is not part of SI-IQ. Arena (Hugging Face)

Computed with SI-IQ method version 1.0.0. Read the full methodology

Data sources: Arena leaderboard dataset on Hugging Face (CC BY 4.0); Berkeley Function Calling Leaderboard, Gorilla project (Apache-2.0); Epoch AI, ‘Capabilities & benchmarking’ (CC BY 4.0); Source: OpenRouter (openrouter.ai), licensed under CC BY 4.0: reuse and republish with attribution to OpenRouter..