Note: Overall leaderboard rankings may not reflect true model quality — individual benchmarks give a clearer picture. ARC-Challenge MMLU GPQA GSM8K Artificial Analysis Intelligence Index v4.1
← Back to leaderboard

Google: Gemma 4 E4B Instruct

23 benchmarks
CompanyGoogle
Country of originUSA
WeightsOpen weights
Parameter count8.0B
Architecturegemma4
VariantReasoning
OriginInstruction tuning
Download Q4_0 GGUF for llama.cpp

Leaderboard position

#833
Overall rank (of 1122 models)
390
Elo rating
23
Benchmarks
Top 74.2%
Percentile
#58
In Google (of 73 models)
Above: T-Tech: t-lite-it-1.0 #832 Below: DeepSeek: DeepSeek R1 Distill Llama 8B #834

Benchmark highlights

Best benchmark
Codeforces ELO 940
Worst benchmark
WikiText-2 (-ppl) -68.255

Nearest competitor from another company

The model from another company with the most similar benchmark scores.

appeared as nearest in 1 of 23 benchmarks

Qwen: qwen3-4b-instruct-2507
1/23 4.3%
Hugging Face Smol Models Research: smollm3-3b
1/23 4.3%
Apple: simplesd-30b-instruct
1/23 4.3%
DeepSeek: deepseek-v3.2-speciale
1/23 4.3%
empero-ai: qwythos-9b-claude-mythos-5-1m
1/23 4.3%
zyphra: zaya1-8b
1/23 4.3%
Tencent: hy3-preview-base
1/23 4.3%
Qwen: qwen3-vl-32b-instruct
1/23 4.3%
Intern Language Models: intern-s1
1/23 4.3%
DeepSeek: DeepSeek R1
1/23 4.3%
Tencent: youtu-llm-2b
1/23 4.3%
PaddlePaddle: PaddleOCR-VL
1/23 4.3%
MWS AI: cotype-nano
1/23 4.3%
Z.ai: GLM 4.6V Flash
1/23 4.3%
DeepSeek: DeepSeek V4 Flash
1/23 4.3%
perplexity: sonar-reasoning
1/23 4.3%
SpaceXAI: Grok 3 Mini
1/23 4.3%
SpaceXAI: Grok 4.1 Fast
1/23 4.3%
OpenBMB: minicpm5-1b
1/23 4.3%
Model Votes % of benchmarks
Qwen: qwen3-4b-instruct-2507 1 4.3%
Hugging Face Smol Models Research: smollm3-3b 1 4.3%
Apple: simplesd-30b-instruct 1 4.3%
DeepSeek: deepseek-v3.2-speciale 1 4.3%
empero-ai: qwythos-9b-claude-mythos-5-1m 1 4.3%
zyphra: zaya1-8b 1 4.3%
Tencent: hy3-preview-base 1 4.3%
Qwen: qwen3-vl-32b-instruct 1 4.3%
Intern Language Models: intern-s1 1 4.3%
DeepSeek: DeepSeek R1 1 4.3%

Benchmarks

Codeforces ELO 940 ask-my-agent bench (Native) 93.6 CowQA 78.7 MMMLU 76.6 MMLU-Pro 69.4 MATH-Vision 59.5 GPQA Diamond 58.6 MMMU-Pro 52.6 LiveCodeBench v6 52 AIME 2026 42.5 TauBench V2 (Average) 42.2 ParseBench 40.5 CoVoST 35.54 MedXPertQA MM 28.7 MRCR v2 8 needle 25.4 Artificial Analysis Intelligence Index v4.0 (Maximum Reasoning) 18.8 Artificial Analysis Intelligence Index v4.1 (Maximum Reasoning) 12 Artificial Analysis Intelligence Index v4.3.2 9 CritPt 1 -(FLEURS) -0.08 -(CoderPPL) -2.6205 AA-Omniscience Index -20 WikiText-2 (-ppl) -68.255

Predicted scores

Estimated from 25 similar models and the global Elo rating. These are projections, not measured results.

Benchmark Predicted Confidence Based on
AIME 2025 ~56.55 65% NVIDIA Corporation: NVIDIA Nemotron 3 Nano 4B, Technology Innovation Institute: falcon-h1r-7b, Qwen: qwen3-14b
MATH-500 ~82.08 64% NVIDIA Corporation: NVIDIA Nemotron 3 Nano 4B, Technology Innovation Institute: falcon-h1r-7b, Qwen: Qwen3 4B
IFBench ~47.22 64% Technology Innovation Institute: falcon-h1r-7b, NVIDIA Corporation: nvidia-nemotron-3.5-bolt-e4b, Meta Llama: Llama 3.3 70B Instruct
MMLU ~54.6 64% DeepSeek: DeepSeek R1, MWS AI: cotype-nano, LiquidAI: lfm-2.5-8b-a1b
Humanity's Last Exam (no tools) ~4.55 63% Technology Innovation Institute: falcon-h1r-7b, Meta Llama: Llama 3.3 70B Instruct, SpaceXAI: Grok 3
IFEval ~76.02 63% Qwen: Qwen3 4B, LiquidAI: lfm-2.5-8b-a1b, OpenBMB: minicpm5-1b
GSM8K ~69.21 63% DeepSeek: deepseek-v3, IBM Granite: granite-4.1-30b, IBM Granite: Granite 4.1 8B
ARC-Easy ~67.82 63% MWS AI: cotype-nano, LiquidAI: lfm-2.5-8b-a1b, OpenBMB: minicpm5-1b
ARC-Challenge ~54.48 63% MWS AI: cotype-nano, LiquidAI: lfm-2.5-8b-a1b, OpenBMB: minicpm5-1b
HellaSwag ~65.5 63% MWS AI: cotype-nano, LiquidAI: lfm-2.5-8b-a1b, OpenBMB: minicpm5-1b
SuperGPQA ~32.91 63% DeepSeek: deepseek-v3, DeepSeek: DeepSeek R1, Meta Llama: Llama 3.3 70B Instruct
WinoGrande ~65.83 63% MWS AI: cotype-nano, LiquidAI: lfm-2.5-8b-a1b, OpenBMB: minicpm5-1b
TruthfulQA ~37.64 63% MWS AI: cotype-nano, LiquidAI: lfm-2.5-8b-a1b, OpenBMB: minicpm5-1b
SWE-bench Verified ~28.29 61% Qwen: Qwen3 4B, OpenAI: gpt-oss-20b, Qwen: Qwen3 8B
SciCode ~21.06 61% Technology Innovation Institute: falcon-h1r-7b, Meta Llama: Llama 3.3 70B Instruct, SpaceXAI: Grok 3
GPQA ~41.53 61% NVIDIA Corporation: NVIDIA Nemotron 3 Nano 4B, Inception: Mercury 2, NVIDIA Corporation: nvidia-nemotron-3-nano-30b-a3b
Artificial Analysis Agentic Index (Maximum Reasoning) ~6.41 61% Technology Innovation Institute: falcon-h1r-7b, DeepSeek: DeepSeek R1, Meta Llama: Llama 3.3 70B Instruct
Text Arena (Snapshot from 20 june 2026) ~1290.85 61% DeepSeek: DeepSeek R1, OpenAI: gpt-oss-20b, Inception: Mercury 2
HMMT Feb 2026 ~58.74 60% Qwen: Qwen3 4B, OpenBMB: minicpm5-1b, Qwen: Qwen3.5 35B A3B
HumanEval ~65.84 60% Qwen: Qwen3 4B, Yandex: YandexGPT Lite 5, Hugging Face Smol Models Research: smollm3-3b
AA-LCR ~36.48 60% Meta Llama: Llama 3.3 70B Instruct, SpaceXAI: Grok 3, SpaceXAI: Grok 3 Mini
MATH ~45.85 60% Mistral AI: ministral-3-14b-instruct-2512, Tencent: hy3-preview-base, Yandex: YandexGPT Lite 5
BBH ~58.68 60% Qwen: Qwen3 4B, OpenBMB: minicpm5-1b, Hugging Face Smol Models Research: smollm3-3b
MMLU-Redux ~73.2 60% Qwen: Qwen3 4B, OpenBMB: minicpm5-1b, Tencent: hy3-preview-base
MBPP ~62.53 59% Qwen: Qwen3 4B, Hugging Face Smol Models Research: smollm3-3b, Tencent: youtu-llm-2b