Gemma 3n E4B Instruct
Jun 2025
Gemma 3n E4B InstructCompare inon
Benchmarks of Gemma 3n E4B Instruct33
BenchmarkByScoreRank
Overall1
Knowledge4
Science2
Coding1
Agents1
Multimodal1
Writing17
Documents1
Legacy · saturated or no longer reported5
No benchmark matches
—per task?
self-host onlyprice
—output speed?
32Kcontext
Rankingin non-thinking mode?
Alternatives?
ModelScoreCostPer taskWhat changes
BetterGPT-5 Minihighwins 64%—$0.053
OpenNVIDIA Nemotron 3.5 Lightning 30B A3B NVFP4wins 59%——Memory (4-bit): 3.9 GB → 11 GB+1BetterGPT-6 Lunamaxwins 71%—$0.068Function calling only with none · Chat Completions+3
BetterClaude Haiku 5.5highwins 72%—$0.079Remove sampling parameters+5BetterGPT-5.6 Lunaxhighwins 74%—$0.085ModelScoreCostPer taskWhat changes
Nothing here yet.
ModelScoreCostPer taskWhat changes
GPT-5 Minihighwins 64%—$0.053GPT-6 Lunamaxwins 71%—$0.068Function calling only with none · Chat Completions+3Claude Haiku 5.5highwins 72%—$0.079Remove sampling parameters+5GPT-5.6 Lunaxhighwins 74%—$0.085GPT-5.6 Terraxhighwins 76%—$0.63Prompts with >272K input tokensModelScoreCostPer taskWhat changes
NVIDIA Nemotron 3.5 Lightning 30B A3B NVFP4Openwins 59%——Memory (4-bit): 3.9 GB → 11 GB+1
Granite 4.2 8BunstatedOpenwins 57%—$0.06 / $0.25Memory (4-bit): 3.9 GB → 4.4 GB+1Gemma 3 12BunstatedOpenwins 56%—$0.05 / $0.10Memory (4-bit): 3.9 GB → 6.1 GBGranite 4.2 3BunstatedOpenwins 53%—$0.03 / $0.12No image input
Molmo2 8BunstatedOpenwins 52%——Memory (4-bit): 3.9 GB → 4.3 GB+1ModelScoreCostPer taskWhat changes
Gemini 4 Argonhighwins 83%—*$1.62Synchronous output stops at 65,536+2Gemini 3.8 Flashhighwins 80%—*$3.49Sampling parameters are ignored+1Gemini 3.6 Flashhighwins 77%—*$1.60Gemini 3.1 Pro Previewunstatedwins 77%—$2 / $12Cannot use both thinking controlsGemini 3 Flash Previewunstatedwins 74%—$0.50 / $3Gemini 2.5 Prounstatedwins 74%—$1.25 / $10Gemma 4 31BunstatedOpenwins 74%—$0.10 / $0.34Memory (4-bit): 3.9 GB → 19 GBGemma 4 26B A4BunstatedOpenwins 72%—$0.15 / $0.60Memory (4-bit): 3.9 GB → 15 GBAll Google models →Run it yourself
estimateFits on✓ 8 GB (phone, laptop) · 8-bit✓ 16 GB · full quality✓ 24 GB GPU · full quality✓ 48 GB GPU · full quality✓ 80 GB GPU · full quality
LicenceGemma termscommercial use with conditions
Benchmarks
all categoriesChat & Knowledge
Humanity's Last Exam4.5#297
AA Omniscience Index-81.3#319
CritPt (physics)0#182
AA Intelligence Index 4.3.2≈4.8#329
GPQA Diamond29.6#300
AIME 2025 · AA archived14.3#120
Coding
Terminal-Bench 2.10.7#152
LiveCodeBench · AA archived14.6#142
Agents & Tool Use
τ³-Banking · AA0.2#134
τ²-bench · AA5#231
Writing & Language
IFBench (instructions)27.9#219
Vision & Documents
MMMU-Pro (vision)26.2#127
FAQ
How much does Gemma 3n E4B Instruct cost?›
There is no per-unit API price: self-host only.
Is Gemma 3n E4B Instruct free to use?›
There is no free API. The weights are open, so you can run it yourself.
How good is Gemma 3n E4B Instruct, and what is it best at?›
#294 of 336 in Chat & Knowledge, #181 of 239 in Coding, #228 of 242 in Agents & Tool Use, #136 of 197 in Writing & Language, #124 of 125 in Vision & Documents.
Is Gemma 3n E4B Instruct better than GPT-4o Mini?›
On Arena Text · instruction following, Gemma 3n E4B Instruct scores 1,255.1 vs 1,258.3 — about the same.
What is the context window of Gemma 3n E4B Instruct?›
32K tokens.
When was Gemma 3n E4B Instruct released? Is there a newer version?›
Released 2025-06-26. It has been superseded by Gemma 4 E4B.
Is Gemma 3n E4B Instruct open source?›
Yes, the weights are open (gemma licence).
Can I run Gemma 3n E4B Instruct locally?›
The weights are open and can be self-hosted; hardware needs are not confirmed yet.
Can Gemma 3n E4B Instruct read images?›
It accepts text and images as input.
What is Gemma 3n E4B Instruct?›
Gemma 3n E4B Instruct is an open-weights language model from Google, released Jun 2025.
What are the best alternatives to Gemma 3n E4B Instruct?›
GPT-5 Mini (better), NVIDIA Nemotron 3.5 Lightning 30B A3B NVFP4 (open weights).