NVIDIAGemma-4-31B-IT
Best configuration ranks #176 of 610 on the BenchLeader Index at 55.4 ±6.1. Last measured 8 Sept 2026. Released 2 Apr 2026.
- Blended price
- $0.205/M
- $0.140 in · $0.400 out
- Output speed
- 35 tok/s
- First answer
- 51 s
- first token 1.10 s
- Context
- 256k
- Overall index55
- Reasoning61
- Coding50
- Agents & tools65
- Knowledge33
- Instruction following73
- Human preference56
- Multimodal60
- Long context61
- Composite48
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 | 50.4 | #285 | 57 tok/s | 1.51 s | $0.0002 | Agents & tools 60 · Composite 47 · Instruction following 54 · Knowledge 40 · Long context 49 · Multimodal 55 |
| minimal | – | – | 35 tok/s | 51 s | – | Maths 56 · Reasoning 56 |
| defaultbest | 55.4 | #176 | 35 tok/s | 51 s | $0.0002 | Agents & tools 65 · Coding 50 · Composite 48 · Human preference 56 · Instruction following 73 · Knowledge 33 · Long context 61 · Multimodal 60 · Reasoning 61 |
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 | minimal | default | Source | Trend |
|---|---|---|---|---|---|
| GPQA Diamond | – | 75.8%#131 | – | Epoch AI Benchmarking Hub | |
| LMArena Hard Prompts | – | – | 1474#65 | LMArena | |
| GPQA Diamond (AA)not in index | 76.3%#230 | – | 85.7%#119 | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 11.8%#226 | – | 23.6%#142 | Artificial Analysis | |
| Kagi LLM Benchmark | – | – | 63.5%#42 | Kagi LLM Benchmark |
Coding
| Benchmark | no reasoning | minimal | default | Source | Trend |
|---|---|---|---|---|---|
| SciCode | – | – | 43.4%#91 | SciCode | |
| WeirdML | – | – | 52.3%#63 | WeirdML | |
| LMArena Coding | – | – | 1499#63 | LMArena | |
| LMArena WebDev | – | – | 1363#83 | LMArena | |
| SciCode (AA)not in index | – | – | 45.5%#90 | Artificial Analysis |
Agents & tools
| Benchmark | no reasoning | minimal | default | Source | Trend |
|---|---|---|---|---|---|
| Terminal-Bench Hard | 30.3%#110 | – | 36.4%#69 | Artificial Analysis | |
| τ²-Bench Telecom (AA)not in index | 65.5%#164 | – | 59.9%#175 | Artificial Analysis |
Maths
| Benchmark | no reasoning | minimal | default | Source | Trend |
|---|---|---|---|---|---|
| OTIS Mock AIME | – | 73.3%#114 | – | Epoch AI Benchmarking Hub |
Knowledge
| Benchmark | no reasoning | minimal | default | Source | Trend |
|---|---|---|---|---|---|
| SimpleQA Verified | – | – | 10.4%#73 | Epoch AI Benchmarking Hub | |
| AA-Omniscience | -51.7#331 | – | -47.9#299 | Artificial Analysis |
Instruction following
| Benchmark | no reasoning | minimal | default | Source | Trend |
|---|---|---|---|---|---|
| IFBench | 53.5%#154 | – | 75.6%#29 | Artificial Analysis |
Human preference
| Benchmark | no reasoning | minimal | default | Source | Trend |
|---|---|---|---|---|---|
| LMArena Text | – | – | 1451#64 | LMArena | |
| EQ-Bench 4 | – | – | 1120#21 | EQ-Bench |
Multimodal
| Benchmark | no reasoning | minimal | default | Source | Trend |
|---|---|---|---|---|---|
| LMArena Vision | – | – | 1276#31 | LMArena | |
| MMMU-Pro | 70.3%#113 | – | 73.4%#95 | Artificial Analysis | |
| MMMU-Pro (official)not in index | – | – | 76.9%#5 | MMMU |
Long context
| Benchmark | no reasoning | minimal | default | Source | Trend |
|---|---|---|---|---|---|
| AA-LCR | 46.7%#262 | – | 69.7%#168 | Artificial Analysis |
Composite
| Benchmark | no reasoning | minimal | default | Source | Trend |
|---|---|---|---|---|---|
| Epoch Capabilities Indexnot in index | – | – | 142.7#83 | Epoch AI Benchmarking Hub | |
| AA Intelligence Index | 13.9#237 | – | 15.4#215 | 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 |
|---|---|---|---|---|---|---|
| Friendli | 93 tok/s | 0.74 s | $0.140 | $0.400 | 262k | – |
| ModelRun [by Modular] | 37 tok/s | 0.23 s | $0.750 | $1.00 | 262k | fp4 |
| CoreWeave | 34 tok/s | 0.76 s | $0.100 | $0.340 | 262k | fp4 |
| DeepInfra Turbo | 29 tok/s | 0.82 s | $0.090 | $0.340 | 262k | fp4 |
| Venice | 27 tok/s | 1.19 s | $0.120 | $0.360 | 256k | bf16 |
| DeepInfra Ultra | 26 tok/s | 2.69 s | $0.270 | $0.760 | 131k | fp8 |
| SiliconFlow | 25 tok/s | 3.87 s | $0.130 | $0.400 | 262k | fp8 |
| NovitaAI | 20 tok/s | 1.01 s | $0.140 | $0.400 | 262k | bf16 |
| DeepInfra | 19 tok/s | 1.44 s | $0.130 | $0.380 | 262k | fp8 |
| Parasail | 19 tok/s | 1.34 s | $0.150 | $0.400 | 262k | fp8 |
| Together | 17 tok/s | 0.68 s | $0.390 | $0.970 | 262k | – |
| Crusoe | 14 tok/s | 0.84 s | $0.140 | $0.400 | 262k | – |
| SambaNova | 14 tok/s | 2.06 s | $0.380 | $1.15 | 262k | – |
| Chutes | 10 tok/s | 4.67 s | $0.120 | $0.370 | 131k | fp4 |
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 | $0.0002 | – | 59.2 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0019 | – | 1.1 min |
| Code edit | 6,000 / 1,500 | $0.0014 | – | 1.6 min |
| Agentic coding session | 60,000 / 4,000 | $0.010 | – | 2.7 min |
| Structured extraction | 2,000 / 200 | $0.0004 | – | 56.4 s |
See also
Data as of 9 Sept 2026. Compare these configurations.