Models / Gemma 3 4B

Google

Gemma 3 4B

Gemma 3 4B is strongest at sticking to the facts; worth a look for simple, high-volume work where cost matters most.

Results 16 Nov 2026 · 29 results · 5 sources

Test Gemma 3 4B on your prompt → It runs beside the models you could use instead, with every answer checked and costed.

Price per million tokens
$0.04 in · $0.05 out
Typical host on OpenRouter, 10 Nov 2026 · compare prices
Speed
Not measured
Intelligence Index
4.8
Artificial Analysis
Spring Prompt overall
Not ranked yet
needs our benchmarks and two groups of results
Developer
Google, based in the US
Weights published
Released
11 Mar 2025
131,072 tokens of context

Where it stands among the models you could choose

Available models grouped by price, intelligence, speed and developer location, plus models with published weights. Retired models are excluded.

AmongIntelligence IndexSpring Prompt overallPrice, blendedSpeed
Models available today 182nd of 185Claude Opus 5.5 – $0.04 blended, 0th percentilecheapest: Mistral Nemo –
Models priced under $0.25 per million tokens 38th of 41Claude Haiku 5.5 – sets this group –
Models with similar intelligence sets this group – $0.04 blended, 0th percentilethe cheapest –
Models from developers based in the US 90th of 92Claude Opus 5.5 – $0.04 blended, 0th percentilethe cheapest –
Open-weights models 101st of 104MiMo-V2.6-Pro – $0.04 blended, 1st percentilecheapest: Mistral Nemo –

The name under each rank is the leader of that group; green is the top quarter of the group and red the bottom quarter, by rank. Price is shown as a percentile: the share of the group that costs less, so lower is cheaper. "Sets this group" marks the measure the group is defined by. Spring Prompt overall is our score out of 100 across every source (how it works). Price is the developer's list price (or the typical OpenRouter host where we haven't read one) for three input tokens to one output; intelligence and speed are from Artificial Analysis (data sourced from Artificial Analysis); "similar intelligence" means within 4 points on its Intelligence Index. Developer locations are where each company is based, not where a model is served. Groups under 3 models are not ranked.

How good is it, and for what?

Each use case ranks the available models its benchmarks measured, then those priced under $0.25 per million tokens. The bar shows its percentile in that field, best to the right. Open a row for the results behind it.

Product listingsTurning a sparse product feed and photos into listings that can go live Not measured
  • CatalogBench: has not measured this model.
Decks from an analysisTurning a finished analysis into a deck you could present as it is Not measured
  • DeckBench: has not measured this model.
User surveysPlanning a user survey and reading its results without being misled Not measured
  • SurveyBench: has not measured this model.
Marketing planningPlanning a year of ad spend without overspending Not measured
  • ROASBench: has not measured this model.
Agents and tool useMulti-step tasks with tools: support desks, coding agents, function calls 158th of 159 available · 36th of 37 at its price
BenchmarkGemma 3 4BRank among availableBest available
Berkeley Function Calling Leaderboard (BFCL) V4BFCL: overall accuracy · reported 19.6% 35th of 36 Claude Opus 4.5 77.5%
Artificial AnalysisArtificial Analysis: Terminal-Bench 2.1 · reported 0.4% 105th of 107 Claude Fable 5.1 91.4%
Artificial AnalysisArtificial Analysis: τ-bench banking · reported 0.4% 104th of 104 Grok 4.6 50.7%
  • tau2-bench: has not measured this model.
  • OpenHands Index: has not measured this model.
  • Microsoft STATE-Bench: has not measured this model.
  • Vending-Bench 2: has not measured this model.
Professional workReal tasks from banking, consulting and law, business documents and freelance projects Not measured
  • APEX-Agents: has not measured this model.
  • GDP.pdf: has not measured this model.
  • Remote Labor Index: has not measured this model.
Reasoning and knowledgeHard questions across science, maths and general knowledge 181st of 187 available · 39th of 41 at its price
BenchmarkGemma 3 4BRank among availableBest available
Artificial AnalysisArtificial Analysis: Artificial Analysis Intelligence Index · reported 4.8 182nd of 185 Claude Opus 5.5 57.6
Artificial AnalysisArtificial Analysis: Humanity's Last Exam · reported 5.3% 147th of 186 Claude Opus 5.5 61.4%
Artificial AnalysisArtificial Analysis: GPQA Diamond · reported 29.1% 165th of 168 GPT-6 Astra 96.3%
WritingWhat people prefer in blind comparisons, and judged writing quality 157th of 169 available · 31st of 39 at its price
BenchmarkGemma 3 4BRank among availableBest available
Arena (formerly LMArena)Arena Text: overall · reported 1,303 136th of 153 Gemini 4 Argon 1,525
UGI LeaderboardUGI: writing score · reported 24.0 118th of 124 Gemini 3.8 Flash 78.6
Sticking to the factsSummarising without inventing things, and factual answers 32nd of 139 available · 10th of 30 at its price
BenchmarkGemma 3 4BRank among availableBest available
Vectara Hallucination LeaderboardVectara: hallucination rate · reported 6.4% 19th of 71 GPT-5.4 nano 3.1%
  • Arena (formerly LMArena): has not measured this model.
  • SimpleQA Verified (Epoch AI): has not measured this model.
Speed under pressureGood decisions against a real clock (fast chess) Not measured
  • BulletBench: has not measured this model.

Nearest alternatives

Against Gemma 3 4B at $0.04 blended, Intelligence Index 4.8. "Similar intelligence" means within 4 points or better.

Where to run it

2 hosts on OpenRouter at the standard service tier, with a median of $0.05 blended per million tokens; quantisation where the host states it: BF16. Cheapest first.

HostQuantisationInputOutputBlended
ParasailBF16$0.04$0.05$0.04
DeepInfraBF16$0.05$0.10$0.06

Prices per million tokens from OpenRouter's public endpoint list, checked 10 Nov 2026.

Against the alternatives

The models you would most likely weigh it against: the leaders of the groups above.

Scroll sideways to see every alternative.

Gemma 3 4BClaude Haiku 5.5
most intelligent at under $0.25 per million tokens
Claude Opus 5.5
most intelligent available today
MiMo-V2.6-Pro
most intelligent with open weights
Price per million tokens $0.04 $0.20$8.00$0.54
Tokens a second – 2409643
Intelligence Index 4.8 43.457.646.3
Spring Prompt overall – 4082–
Rank among available models, by use case
Product listings – 20th13th–
Decks from an analysis – 18th7th–
User surveys – 7th4th–
Marketing planning – 30th9th–
Agents and tool use 158th of 159 28th3rd23rd
Professional work – –3rd–
Reasoning and knowledge 181st of 187 25th1st13th
Writing 157th of 169 –5th23rd
Sticking to the facts 32nd of 139 –3rd10th
Speed under pressure – 21st––

Shaded figures are better than Gemma 3 4B's. A dash means no result.

Where it has been measured

Cheapest of the 104 models from developers based in the US, and cheapest of the 39 models with similar intelligence. 38th most intelligent of the 41 models priced under $0.25 per million tokens. Among models available today, in the bottom quarter for agents and tool use, reasoning and knowledge, and writing.

Artificial Analysis lists it but has not published a speed for it yet

Every result

29 published results from 5 sources, each in the source's own units, with the configuration that produced it.

Arena (formerly LMArena) · reported by Arena (formerly LMArena) · 6 results
MeasureValueRankConfigurationDated
Arena Text: business, management and financeArena rating, higher is better 1,330
1,300–1,360
233 of 406 Gemma 3 4B 2 Oct 2026
Arena Text: creative writingArena rating, higher is better 1,275
1,252–1,297
269 of 411 Gemma 3 4B 2 Oct 2026
Arena Text: expert promptsArena rating, higher is better 1,260
1,219–1,301
295 of 363 Gemma 3 4B 2 Oct 2026
Arena Text: instruction followingArena rating, higher is better 1,267
1,251–1,284
296 of 413 Gemma 3 4B 2 Oct 2026
Arena Text: overallArena rating, higher is better 1,303
1,294–1,313
281 of 413 Gemma 3 4B 2 Oct 2026
Arena Text: writing, literature and languageArena rating, higher is better 1,298
1,281–1,315
263 of 412 Gemma 3 4B 2 Oct 2026

Not shown: Not accuracy or correctness: it ranks which answer voters preferred, with answer length and formatting controlled for.

Artificial Analysis · reported by Artificial Analysis · 10 results
MeasureValueRankConfigurationDated
Artificial Analysis: Artificial Analysis Coding Indexindex score, higher is better 2.7 184 of 187 Gemma 3 4B 20 Oct 2026
Artificial Analysis: Artificial Analysis Intelligence Indexindex score, higher is better 4.8 410 of 413 Gemma 3 4B 20 Oct 2026
Artificial Analysis: GPQA Diamond% of questions, higher is better 29.1% 354 of 360 Gemma 3 4B 20 Oct 2026
Artificial Analysis: Humanity's Last Exam% of questions, higher is better 5.3% 337 of 419 Gemma 3 4B 20 Oct 2026
Artificial Analysis: IFBench% of instructions, higher is better 28.3% 277 of 288 Gemma 3 4B 20 Oct 2026
Artificial Analysis: Long-context reasoning (AA-LCR)% of questions, higher is better 6.7% 386 of 407 Gemma 3 4B 20 Oct 2026
Artificial Analysis: Terminal-Bench 2.1% of tasks, higher is better 0.4% 182 of 188 Gemma 3 4B 20 Oct 2026
Artificial Analysis: Terminal-Bench Hard% of tasks, higher is better 0.8% 271 of 282 Gemma 3 4B 20 Oct 2026
Artificial Analysis: τ-bench banking% of tasks, higher is better 0.4% 175 of 175 Gemma 3 4B 20 Oct 2026
Artificial Analysis: τ²-bench telecom% of tasks, higher is better 5.0% 280 of 286 Gemma 3 4B 20 Oct 2026

Not shown: Not your codebase or tools.

Berkeley Function Calling Leaderboard (BFCL) V4 · reported by Berkeley Function Calling Leaderboard (BFCL) V4 · 8 results
MeasureValueRankConfigurationDated
BFCL: irrelevance detection% correct, higher is better 53.9% 96 of 109 Gemma 3 4B (bfcl prompt) 16 Dec 2025
BFCL: memory% correct, higher is better 8.6% 75 of 109 Gemma 3 4B (bfcl prompt) 16 Dec 2025
BFCL: multi-turn tasks% correct, higher is better 0.4% 103 of 109 Gemma 3 4B (bfcl prompt) 16 Dec 2025
BFCL: overall accuracy% correct, higher is better 19.6% 101 of 109 Gemma 3 4B (bfcl prompt) 16 Dec 2025
BFCL: relevance detection% correct, higher is better 100.0% 1 of 109 Gemma 3 4B (bfcl prompt) 16 Dec 2025
BFCL: single-turn calls (curated)% correct, higher is better 61.1% 96 of 109 Gemma 3 4B (bfcl prompt) 16 Dec 2025
BFCL: single-turn calls (user-contributed)% correct, higher is better 60.8% 83 of 109 Gemma 3 4B (bfcl prompt) 16 Dec 2025
BFCL: web search% correct, higher is better 1.0% 81 of 109 Gemma 3 4B (bfcl prompt) 16 Dec 2025

Not shown: Not general refusal or safety behaviour.

UGI Leaderboard · reported by UGI Leaderboard · 3 results
MeasureValueRankConfigurationDated
UGI: requested-length error% off the requested word count, lower is better 9.0% 75 of 379 Gemma 3 4B 12 Sep 2025
UGI: style adherencescore from 0 to 1, higher is better 0.29 354 of 379 Gemma 3 4B 12 Sep 2025
UGI: writing scorescore out of 100, higher is better 24.0 320 of 379 Gemma 3 4B 12 Sep 2025

Not shown: Not other format limits such as character counts or bullet counts.

Vectara Hallucination Leaderboard · reported by Vectara Hallucination Leaderboard · 2 results
MeasureValueRankConfigurationDated
Vectara: answer rate% of documents, higher is better 67.3% 107 of 108 Gemma 3 4B 22 Sep 2026
Vectara: hallucination rate% of summaries, lower is better 6.4% 28 of 108 Gemma 3 4B 22 Sep 2026

Not shown: Not a quality score: a low rate usually means content filters were triggered, and hallucination rates are measured on answered documents only.

Will Gemma 3 4B work for your product?

Benchmarks aren't your task. Paste the prompt your product uses and run it beside the models you could use instead, with every answer checked and costed.

New here? Your email starts a free trial. No credit card needed.