gemma-3-4b-it: benchmark scores, price and ranking
gemma-3-4b-it from Google DeepMind ranks #734 of 743 LLMs on the LLMs Tiger aggregate benchmark score (39.9), based on 5 benchmark results captured up to 2026-10-13.
- Organisation
- Google DeepMind
- Type
- Open-weight / local model
- Open weights
- Yes
- Released
- 2025-03-12 (source: epoch)
- Cheapest API price
- $0.050 in / $0.100 out per million tokens (via deepinfra, observed 2026-10-13T20:47Z)
- Aggregate rank
- #734 of 743 (score 39.9)
Relative to other models, gemma-3-4b-it is strongest on OTIS Mock AIME (#144 of 189) and weakest on LMCA (#108 of 108).
Across the 5 benchmarks where at least ten models are scored, gemma-3-4b-it is in the top 10% on 0 and in the top half on 0.
Its cheapest listed input price, $0.050 per million tokens, is lower than that of 90% of the 427 models with a listed price.
gemma-3-4b-it benchmark results
| Benchmark | Score | Rank | Source | Captured |
|---|---|---|---|---|
| OTIS Mock AIME | 7.5% | #144 of 189 | Epoch AI Benchmarking Hub | 2026-10-13 |
| Chess Puzzles | 0.0% | #104 of 122 | Epoch AI Benchmarking Hub | 2026-10-13 |
| DTBench | 50.9% | #128 of 146 | Epoch AI Benchmarking Hub | 2026-10-13 |
| GPQA Diamond | 23.2% | #268 of 275 | Epoch AI Benchmarking Hub | 2026-10-13 |
| LMCA | 2.8% | #108 of 108 | Epoch AI Benchmarking Hub | 2026-10-13 |
About gemma-3-4b-it
gemma-3-4b-it is an open-weight model that can run on your own hardware from Google DeepMind. It was released on 2025-03-12, according to Epoch AI. The cheapest API listing we track charges $0.050 per million input tokens and $0.100 per million output tokens, through deepinfra. We hold 5 benchmark results for it across 3 areas, and none is in the top tenth of its board.
Where gemma-3-4b-it stands
On our aggregate score, gemma-3-4b-it is #734 of 743 ranked models, in the lower half. Its score is 39.9; the scale is centred on 50, so it sits 10.1 points below the typical model. That beats 1% of the ranked field. Among the 41 ranked Google DeepMind models it is #39. The next one up is gemini-1.0-pro-001 at 41.5. The next one down is gemma-3-27b-it at 39.7.
gemma-3-4b-it, benchmark by benchmark
gemma-3-4b-it lands below the middle on OTIS Mock AIME: 7.5%, #144 of 189, with 76% of the field scoring higher. Claude Fable 5.1 (max) leads at 100.0%; the closest result above is mistral-large-2411 at 7.8%. It covers competition maths (AIME 2024-25 style). The board median is 57.8%, so gemma-3-4b-it is 50.3 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
Chess Puzzles is a weak spot. gemma-3-4b-it scores 0.0%, #104 of 122, behind 84% of the models measured. The leader, GPT-6 Astra (max), reaches 72.0%. It covers chess tactics puzzles. The board median is 12.5%, so gemma-3-4b-it is 12.5 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
DTBench is a weak spot. gemma-3-4b-it scores 50.9%, #128 of 146, behind 87% of the models measured. The leader, Claude Opus 5.5 (max), reaches 98.9%. It covers dynamic tool-use benchmark. The board median is 68.8%, so gemma-3-4b-it is 17.9 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
GPQA Diamond is a weak spot. gemma-3-4b-it scores 23.2%, #268 of 275, behind 97% of the models measured. The leader, GPT-6 Astra (max), reaches 95.8%. It covers expert-level science questions. The board median is 71.7%, so gemma-3-4b-it is 48.5 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
LMCA is a weak spot. gemma-3-4b-it scores 2.8%, #108 of 108, behind 99% of the models measured. The leader, Claude Opus 5.5 (max), reaches 68.2%. It covers language-model capability assessment. The board median is 29.2%, so gemma-3-4b-it is 26.4 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
Strengths and weaknesses
Reasoning, maths and science: weak, ahead of 13% of the field on average across 2 benchmarks. Its best board here is OTIS Mock AIME (#144 of 189) and its weakest is GPQA Diamond (#268 of 275). General knowledge and chat: weak, ahead of 7% of the field on average across 2 benchmarks. Its best board here is Chess Puzzles (#104 of 122) and its weakest is LMCA (#108 of 108).
Price and value
The median input price across the 427 priced models is $0.500 per million tokens, so gemma-3-4b-it costs less than the typical listing. Models that score within 1.5 points or better and cost less: Qwen2.5-Coder-7B-Instruct ($0.010, 49.5), DeepSeek V4.1 Flash (high) ($0.010, 48.9), Qwen2.5-Coder-3B-Instruct ($0.010, 46.5). Similar scores cost more on gemma-3-27b-it ($0.080) and GPT-4.1 nano ($0.100).
Release history
The Google DeepMind model released before it was gemma-3-27b-it on 2025-03-12, which scores 39.7. The next one was gemini-2.5-pro_32K on 2025-03-25, at 49.2.
Reading these numbers
The 5 scores on this page come from 1 source: Epoch AI Benchmarking Hub (5). No official leaderboard supplied any of them, so treat them as third-party measurements. With 5 results the aggregate rank can move a lot when one new score arrives.
Models ranked nearby
- #731 Llama-3.3-70B-Instruct (40.7)
- #732 claude-2.1 (40.6)
- #733 Llama-4-Scout-17B-16E-Instruct (40.1)
- #735 qwen2.5-7b-instruct (39.9)
- #736 claude-3-haiku-20240307 (39.9)
- #737 gemma-3-27b-it (39.7)
More from Google DeepMind
- Gemini 3.8 Flash (high) (#26, 54.0)
- Gemini 3.7 Flash (high) (#27, 53.9)
- Gemini 3 Pro Preview (#31, 53.7)
- Gemini 3 Deep Think (#60, 52.6)
- AI co-mathematician (#69, 52.3)
- Gemini 3 Flash (#120, 51.5)
Compare it on the live leaderboard → · How the aggregate score works