GoogleGemma 4 12B
Best configuration ranks #258 of 610 on the BenchLeader Index at 52.1 ±8.0. Released 3 Jun 2026.
- Blended price
- $0.150/M
- $0.100 in · $0.300 out
- Output speed
- 107 tok/s
- First answer
- 21 s
- Context
- 256k
- Overall index52
- Agents & tools49
- Knowledge39
- Instruction following71
- Multimodal54
- Long context58
- Composite47
Reasoning-effort configurations
The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Default” means the publisher did not say which setting was used.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| no reasoning | 45.4 | #390 | 104 tok/s | 2.43 s | $0.0001 | Agents & tools 43 · Composite 41 · Instruction following 47 · Knowledge 39 · Long context 43 · Multimodal 46 |
| defaultbest | 52.1 | #258 | 107 tok/s | 21 s | $0.0001 | Agents & tools 49 · Composite 47 · Instruction following 71 · Knowledge 39 · Long context 58 · Multimodal 54 |
Benchmark results
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
| Benchmark | no reasoning | default | Source |
|---|---|---|---|
| GPQA Diamond (AA)not in index | 66.1%#320 | 75.3%#240 | Artificial Analysis |
| Humanity's Last Exam (AA)not in index | 6.3%#322 | 15.7%#192 | Artificial Analysis |
Agents & tools
| Benchmark | no reasoning | default | Source |
|---|---|---|---|
| Terminal-Bench Hard | 11.4%#207 | 18.2%#166 | Artificial Analysis |
| τ²-Bench Telecom (AA)not in index | 31.9%#243 | 36.3%#227 | Artificial Analysis |
Knowledge
| Benchmark | no reasoning | default | Source |
|---|---|---|---|
| AA-Omniscience | -52.8#340 | -52.7#339 | Artificial Analysis |
Instruction following
| Benchmark | no reasoning | default | Source |
|---|---|---|---|
| IFBench | 45.2%#203 | 73.5%#39 | Artificial Analysis |
Multimodal
| Benchmark | no reasoning | default | Source |
|---|---|---|---|
| MMMU-Pro | 62.0%#168 | 69.7%#120 | Artificial Analysis |
Long context
| Benchmark | no reasoning | default | Source |
|---|---|---|---|
| AA-LCR | 35.0%#303 | 63.7%#202 | Artificial Analysis |
Composite
| Benchmark | no reasoning | default | Source |
|---|---|---|---|
| AA Intelligence Index | 9.4#331 | 14.2#233 | Artificial Analysis |
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.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0001 | – | 23.9 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0014 | – | 26.7 s |
| Code edit | 6,000 / 1,500 | $0.0011 | – | 35.1 s |
| Agentic coding session | 60,000 / 4,000 | $0.0072 | – | 58.5 s |
| Structured extraction | 2,000 / 200 | $0.0003 | – | 22.9 s |
See also
Data as of 9 Sept 2026. Compare these configurations.