NVIDIAGemma 3 4B IT
Best configuration ranks #600 of 610 on the BenchLeader Index at 34.5 ±4.5. Last measured 2 Sept 2026. Released 12 Mar 2025.
- Blended price
- $0.063/M
- $0.050 in · $0.100 out
- Output speed
- 27 tok/s
- First answer
- 0.62 s
- Context
- 131k
- Overall index35
- Reasoning27
- Coding39
- Agents & tools34
- Maths27
- Knowledge25
- Instruction following32
- Human preference47
- Multimodal18
- Long context28
- Composite35
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 | default | Source |
|---|---|---|
| GPQA Diamond | 23.2%#266 | Epoch AI Benchmarking Hub |
| LMArena Hard Prompts | 1284#244 | LMArena |
| GPQA Diamond (AA)not in index | 29.1%#495 | Artificial Analysis |
| Humanity's Last Exam (AA)not in index | 5.3%#359 | Artificial Analysis |
| Kagi LLM Benchmark | 25.2%#128 | Kagi LLM Benchmark |
Coding
| Benchmark | default | Source |
|---|---|---|
| LMArena Coding | 1274#269 | LMArena |
Agents & tools
| Benchmark | default | Source |
|---|---|---|
| Terminal-Bench Hard | 0.8%#332 | Artificial Analysis |
| τ²-Bench Telecom (AA)not in index | 5.0%#372 | Artificial Analysis |
| BFCL Overall | 19.6%#76 | Berkeley Function Calling Leaderboard |
Maths
| Benchmark | default | Source |
|---|---|---|
| OTIS Mock AIME | 7.5%#216 | Epoch AI Benchmarking Hub |
Knowledge
| Benchmark | default | Source |
|---|---|---|
| AA-Omniscience | -82.9#466 | Artificial Analysis |
Instruction following
| Benchmark | default | Source |
|---|---|---|
| IFBench | 28.3%#355 | Artificial Analysis |
Human preference
| Benchmark | default | Source |
|---|---|---|
| LMArena Text | 1303#230 | LMArena |
Multimodal
| Benchmark | default | Source |
|---|---|---|
| MMMU-Pro | 29.9%#239 | Artificial Analysis |
| MMMU (validation) | 48.8%#43 | MMMU |
Long context
| Benchmark | default | Source |
|---|---|---|
| AA-LCR | 6.7%#407 | Artificial Analysis |
Composite
| Benchmark | default | Source |
|---|---|---|
| Epoch Capabilities Indexnot in index | 116.0#180 | Epoch AI Benchmarking Hub |
| AA Intelligence Index | 4.8#527 | Artificial 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.
| Provider | Speed | First token | Input $/M | Output $/M | Context | Quantisation |
|---|---|---|---|---|---|---|
| DeepInfra | 27 tok/s | 0.62 s | $0.050 | $0.100 | 131k | bf16 |
Price history
Listed price per 1M tokens over time, as recorded by OpenRouter for the provider with the longest history.
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 | – | 11.7 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0007 | – | 22.8 s |
| Code edit | 6,000 / 1,500 | $0.0005 | – | 56.2 s |
| Agentic coding session | 60,000 / 4,000 | $0.0034 | – | 2.5 min |
| Structured extraction | 2,000 / 200 | $0.0001 | – | 8.0 s |
See also
Data as of 9 Sept 2026. Compare with another model.