BenchLeader
Menu
Best LLMsChoose a modelLeaderboardsModelsBenchmarksProvidersReleases12
GoogleOpen weights: google/gemma-3-4b-itCompare with another model

Gemma 3 4B

Gemma 3 4B is a Google open-weights model, released 12 Mar 2025. Its best configuration ranks #769 of 785 on the BenchLeader Index at 34.5 ±4.1, in the lower half. The ± is the point: 98 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in human preference (46) and lowest in multimodal (17). At $0.063 per million tokens blended it is among the cheapest fifth of ranked models. Output speed of 112 tokens per second puts it faster than most, with a first token in 0.7 s. Last measured 22 Oct 2026.

Blended price
$0.063/M
$0.050 in · $0.100 out
Output speed
112 tok/s
OpenRouter traffic, 7-day median; not yet measured by Artificial Analysis
First answer
0.70 s
Context
131k
Full answer
–
Cost per run
–
Released
12 Mar 2025
How it scores by categoryDashed line = average model (50). One step of 15 = one standard deviation.
  1. Overall index35
  2. Reasoning32
  3. Coding38
  4. Agents & tools31
  5. Maths37
  6. Knowledge24
  7. Instruction following33
  8. Human preference46
  9. Multimodal17
  10. Long context27
  11. Composite35

How its index has moved

8 Sept 2026 to 23 Oct 2026
303744

The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.

Benchmark results

See all LLM benchmarks. One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.

Reasoning

Benchmarknot statedSourceTrend
GPQA Diamond23.2%#287Epoch AI Benchmarking Hub
Arena Hard Prompts1284#268Arena (LMArena)
GPQA Diamond (AA)not in index29.1%#528Artificial Analysis
Humanity's Last Exam (AA)not in index5.3%#436Artificial Analysis
Kagi LLM Benchmark25.2%#129Kagi LLM Benchmark
CritPt0.0%#315Artificial Analysis
LMCAnot in index2.8%#192Epoch AI Benchmarking Hub
DTBenchnot in index50.9%#211Epoch AI Benchmarking Hub

Coding

Benchmarknot statedSourceTrend
Arena Coding1273#294Arena (LMArena)
Terminal-Bench 2.1 (AA)not in index0.4%#228Artificial Analysis

Agents & tools

Benchmarknot statedSourceTrend
Terminal-Bench Hard0.8%#349Artificial Analysis
τ²-Bench Telecom (AA)not in index5.0%#390Artificial Analysis
BFCL Overall19.6%#76Berkeley Function Calling Leaderboard
τ³-Banking (AA)0.4%#203Artificial Analysis
BrowseCompnot in index0.4%#32OpenAI

Maths

Benchmarknot statedSourceTrend
OTIS Mock AIME7.5%#236Epoch AI Benchmarking Hub
Arena Maths1254#265Arena (LMArena)

Knowledge

Benchmarknot statedSourceTrend
AA-Omniscience-82.9#546Artificial Analysis
AA-Omniscience: accuracynot in index7.7%#521Artificial Analysis
AA-Omniscience: non-hallucinationnot in index1.8%#550Artificial Analysis

Instruction following

Benchmarknot statedSourceTrend
IFBench28.3%#374Artificial Analysis
Arena Instruction Followingnot in index1267#268Arena (LMArena)

Human preference

Benchmarknot statedSourceTrend
Arena Text1303#253Arena (LMArena)
Arena Creative Writingnot in index1274#243Arena (LMArena)
Arena Multi-turnnot in index1272#266Arena (LMArena)
Arena Longer Queriesnot in index1312#247Arena (LMArena)

Multimodal

Benchmarknot statedSourceTrend
MMMU-Pro29.9%#275Artificial Analysis
MMMU (validation)48.8%#46MMMU

Long context

Benchmarknot statedSourceTrend
AA-LCR6.7%#480Artificial Analysis

Composite

Benchmarknot statedSourceTrend
Epoch Capabilities Indexnot in index115.9#192Epoch AI Benchmarking Hub
AA Intelligence Index v4.3.24.8#618Artificial Analysis

Where to run it

Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.

ProviderSpeedFirst tokenInput $/MOutput $/MContextQuantisation
Parasail127 tok/s0.84 s$0.040$0.050131kbf16
DeepInfra83 tok/s0.81 s$0.050$0.100131kbf16

What a task costs

Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache. Reasoning tokens are not modelled.

WorkloadTokens in / outCostWith cachingTime
Chat reply400 / 300$0.0001–3.4 s
Summarize a 30-page report12,000 / 600$0.0007–6.1 s
Code edit6,000 / 1,500$0.0005–14.1 s
Agentic coding session60,000 / 4,000$0.0034–36.4 s
Structured extraction2,000 / 200$0.0001–2.5 s

More from Google

New this month

See also

Best under $2/MBest under $0.50/MBest open weightsBest for agentsAll Google models

Questions people ask

What are Gemma 3 4B's benchmark scores?
As of 24 Oct 2026, Gemma 3 4B scores 34.5 ±4.1 on the BenchLeader Index, strongest in human preference (46) and weakest in multimodal (17). That rests on 17 independent benchmark results, and every score on this page links to its publisher.
How does Gemma 3 4B rank?
Gemma 3 4B is #769 of 785 ranked configurations on 24 Oct 2026. 98 other configurations score within its margin, so the exact position matters less than the band. The leader is Claude Opus 5.5 (max effort) at 72.6.
How much does Gemma 3 4B cost per 1M tokens?
As of 24 Oct 2026, Gemma 3 4B costs $0.050 per 1M input tokens and $0.100 per 1M output tokens, $0.063 blended (three input tokens per output token).
When was Gemma 3 4B released?
Google released Gemma 3 4B on 12 Mar 2025.

Data as of 24 Oct 2026. Compare with another model.

Cite as: BenchLeader, “Gemma 3 4B: benchmarks, pricing, speed and rank”, data as of 24 Oct 2026, https://www.benchleader.com/models/gemma-3-4b.