GoogleGemma 4 31B
Gemma 4 31B is a Google open-weights model, released 2 Apr 2026. Its best configuration (thinking reasoning effort) ranks #329 of 785 on the BenchLeader Index at 52.0 ±8.0, in the upper half. The ± is the point: 443 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 instruction following (74) and lowest in knowledge (40). At $0.250 per million tokens blended it is cheaper than most ranked models. Output speed of 36 tokens per second puts it in the slowest quarter, with a first answer in 49.4 s. It has been measured at 5 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 2 Apr 2026.stale: no new result in six months
- Blended price
- $0.250/M
- $0.200 in · $0.400 out
- Output speed
- 36 tok/s
- measured by Artificial Analysis
- First answer
- 49 s
- first token 0.66 s
- Context
- 262k
- Full answer
- 63 s
- median, reasoning included
- Cost per run
- –
- Released
- 2 Apr 2026
- Overall index52
- Reasoning43
- Agents & tools50
- Knowledge40
- Instruction following74
- Multimodal56
- Long context60
- Composite46
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. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| no reasoning | 47.3 | #458 | 52 tok/s | 1.70 s | $0.0002 | Agents & tools 44 · Composite 45 · Instruction following 54 · Knowledge 38 · Long context 48 · Multimodal 53 · Reasoning 41 |
| minimal | – | – | 39 tok/s | 0.66 s | $0.0002 | Maths 56 · Reasoning 55 |
| high | – | – | 39 tok/s | 0.66 s | $0.0002 | |
| thinkingbest | 52.0 | #329 | 36 tok/s | 49 s | $0.0002 | Agents & tools 50 · Composite 46 · Instruction following 74 · Knowledge 40 · Long context 60 · Multimodal 56 · Reasoning 43 |
| not stated | 48.4 | #427 | 39 tok/s | 0.66 s | $0.0002 | Coding 49 · Composite 20 · Human preference 55 · Knowledge 24 · Maths 65 · Multimodal 62 · Reasoning 61 |
How its index has moved
8 Sept 2026 to 23 Oct 2026- thinking
- not stated
- no reasoning
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
| Benchmark | no reasoning | minimal | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|---|
| GPQA Diamond | – | 75.8%#144 | – | – | Epoch AI Benchmarking Hub | |
| Arena Hard Promptseffort not stated | – | – | – | 1476#72 | Arena (LMArena) | |
| LiveBench 2.0 Reasoningnot in indexeffort not stated | – | – | – | 68.3%#50 | LiveBench | |
| GPQA Diamond (AA)not in index | 76.3%#238 | – | 85.7%#120 | – | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 11.8%#289 | – | 23.6%#193 | – | Artificial Analysis | |
| Kagi LLM Benchmarkeffort not stated | – | – | – | 63.5%#42 | Kagi LLM Benchmark | |
| CritPt | 0.0%#315 | – | 1.4%#211 | – | Artificial Analysis | |
| LMCAnot in indexeffort not stated | – | – | – | 39.3%#110 | Epoch AI Benchmarking Hub | |
| DTBenchnot in indexeffort not stated | – | – | – | 82.7%#117 | Epoch AI Benchmarking Hub | |
| ForecastBenchnot in indexeffort not stated | – | – | – | 61.3%#66 | ForecastBench |
Coding
| Benchmark | no reasoning | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| SciCodeeffort not stated | – | – | 45.5%#138 | SciCode | |
| WeirdMLeffort not stated | – | – | 52.3%#66 | WeirdML | |
| Arena Codingeffort not stated | – | – | 1502#73 | Arena (LMArena) | |
| Arena WebDeveffort not stated | – | – | 1368#104 | Arena (LMArena) | |
| LiveBench 2.0 Codingnot in indexeffort not stated | – | – | 49.0%#48 | LiveBench | |
| SciCode (AA)not in index | – | 45.5%#143 | – | Artificial Analysis | |
| Surface Evolver Benchnot in indexeffort not stated | – | – | 30.6%#24 | Epoch AI Benchmarking Hub | |
| ALE-Benchnot in indexeffort not stated | – | – | 925.5#63 | Epoch AI Benchmarking Hub | |
| Terminal-Bench 4.0 (AA)not in index | – | 0.0%#152 | – | Artificial Analysis | |
| Terminal-Bench 2.1 (AA)not in index | 29.2%#157 | 43.5%#133 | – | Artificial Analysis |
Agents & tools
| Benchmark | no reasoning | high | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|---|
| LiveBench Long Agenticnot in indexeffort not stated | – | – | – | 13.3%#46 | LiveBench | |
| LiveBench Browsernot in indexeffort not stated | – | – | – | 4.0%#45 | LiveBench | |
| Terminal-Bench Hard | 30.3%#112 | – | 36.4%#71 | – | Artificial Analysis | |
| τ²-Bench Telecom (AA)not in index | 65.5%#168 | – | 59.9%#180 | – | Artificial Analysis | |
| GDPval-AA v2.1not in index | 2.4%#120 | – | 5.9%#115 | – | Artificial Analysis | |
| τ³-Banking (AA) | 8.9%#153 | – | 14.8%#120 | – | Artificial Analysis | |
| ITBench SRE (AA)not in index | – | – | 37.3%#25 | – | Artificial Analysis | |
| Terminal-Bench 2.0 (Vals)not in index | – | 39.3%#35 | – | – | Vals AI | |
| AutomationBenchnot in index | – | – | 4.9%#166 | – | Artificial Analysis | |
| GDP.pdf (AA)not in index | – | – | 6.0%#154 | – | Artificial Analysis | |
| EnterpriseOps-Gymnot in index | – | – | 28.4%#44 | – | Artificial Analysis |
Maths
| Benchmark | minimal | not stated | Source | Trend |
|---|---|---|---|---|
| OTIS Mock AIME | 73.3%#129 | – | Epoch AI Benchmarking Hub | |
| Arena Mathseffort not stated | – | 1469#52 | Arena (LMArena) |
Knowledge
| Benchmark | no reasoning | high | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|---|
| SimpleQA Verifiedeffort not stated | – | – | – | 10.4%#85 | Epoch AI Benchmarking Hub | |
| LiveBench 2.0 Data Analysisnot in indexeffort not stated | – | – | – | 53.3%#50 | LiveBench | |
| LiveBench Researchnot in indexeffort not stated | – | – | – | 44.4%#47 | LiveBench | |
| AA-Omniscience | -51.7#404 | – | -47.9#370 | – | Artificial Analysis | |
| CaseLaw v2not in index | – | 52.6%#42 | – | – | Vals AI | |
| AA-Omniscience: accuracynot in index | 16.6%#385 | – | 20.0%#307 | – | Artificial Analysis | |
| AA-Omniscience: non-hallucinationnot in index | 18.1%#304 | – | 15.0%#340 | – | Artificial Analysis |
Instruction following
| Benchmark | no reasoning | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| IFBench | 53.5%#157 | 75.6%#29 | – | Artificial Analysis | |
| Arena Instruction Followingnot in indexeffort not stated | – | – | 1454#59 | Arena (LMArena) |
Human preference
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Arena Text | 1452#74 | Arena (LMArena) | |
| EQ-Bench 4 | 1114#24 | EQ-Bench | |
| Arena Creative Writingnot in index | 1419#83 | Arena (LMArena) | |
| Arena Multi-turnnot in index | 1465#72 | Arena (LMArena) | |
| Arena Longer Queriesnot in index | 1470#61 | Arena (LMArena) |
Multimodal
| Benchmark | no reasoning | high | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|---|
| Arena Visioneffort not stated | – | – | – | 1277#39 | Arena (LMArena) | |
| LiveBench Designnot in indexeffort not stated | – | – | – | 58.4%#50 | LiveBench | |
| MMMU-Pro | 70.3%#144 | – | 73.4%#123 | – | Artificial Analysis | |
| MMMU-Pro (official)not in indexeffort not stated | – | – | – | 76.9%#5 | MMMU | |
| MortgageTaxnot in index | – | 61.4%#58 | – | – | Vals AI | |
| SAGEnot in index | – | 55.0%#2 | – | – | Vals AI |
Long context
| Benchmark | no reasoning | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| AA-LCR | 46.7%#331 | 69.7%#222 | – | Artificial Analysis | |
| Arena Documentnot in indexeffort not stated | – | – | 1445#34 | Arena (LMArena) |
Composite
| Benchmark | no reasoning | thinking | not stated | Source | Trend |
|---|---|---|---|---|---|
| Epoch Capabilities Indexnot in indexeffort not stated | – | – | 143.0#96 | Epoch AI Benchmarking Hub | |
| LiveBench 2.0effort not stated | – | – | 41.5%#50 | LiveBench | |
| AA Intelligence Index v4.3.2 | 13.9#304 | 14.7#292 | – | 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 |
|---|---|---|---|---|---|---|
| ModelRun [by Modular] | 198 tok/s | 0.21 s | $0.750 | $1.00 | 262k | fp4 |
| SambaNova | 173 tok/s | 1.36 s | $0.380 | $1.15 | 131k | – |
| Friendli | 116 tok/s | 0.84 s | $0.140 | $0.400 | 262k | – |
| Crusoe | 54 tok/s | 0.39 s | $0.140 | $0.400 | 262k | bf16 |
| CoreWeave | 41 tok/s | 0.66 s | $0.100 | $0.340 | 262k | fp4 |
| DeepInfra (fp4) | 39 tok/s | 0.69 s | $0.135 | $0.510 | 262k | fp4 |
| Venice | 35 tok/s | 0.70 s | $0.120 | $0.360 | 256k | fp4 |
| io.net | 30 tok/s | 0.95 s | $0.361 | $1.09 | 262k | – |
| DeepInfra (fp8) | 29 tok/s | 0.95 s | $0.200 | $0.400 | 262k | fp8 |
| DeepInfra (fp8) | 24 tok/s | 0.82 s | $0.300 | $0.600 | 262k | fp8 |
| DekaLLM | 23 tok/s | 2.62 s | $0.140 | $0.400 | 262k | – |
| SiliconFlow | 22 tok/s | 1.18 s | $0.750 | $1.00 | 262k | fp8 |
| NovitaAI | 21 tok/s | 0.75 s | $0.140 | $0.400 | 262k | bf16 |
| Parasail | 20 tok/s | 1.88 s | $0.150 | $0.400 | 262k | fp8 |
| Chutes | 17 tok/s | 4.09 s | $0.120 | $0.370 | 131k | fp4 |
Price history
The price above, per 1M tokens, every day it changed.
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 at $0.250 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0002 | $0.0002 | 57.7 s |
| Summarize a 30-page report | 12,000 / 600 | $0.0026 | $0.0031 | 1.1 min |
| Code edit | 6,000 / 1,500 | $0.0018 | $0.0020 | 1.5 min |
| Agentic coding session | 60,000 / 4,000 | $0.014 | $0.016 | 2.7 min |
| Structured extraction | 2,000 / 200 | $0.0005 | $0.0006 | 54.9 s |
More from Google
See also
Questions people ask
- What are Gemma 4 31B's benchmark scores?
- As of 25 Oct 2026, Gemma 4 31B (thinking effort) scores 52.0 ±8.0 on the BenchLeader Index, strongest in instruction following (74) and weakest in knowledge (40). That rests on 8 independent benchmark results, and every score on this page links to its publisher.
- How does Gemma 4 31B rank?
- Gemma 4 31B (thinking effort) is #329 of 785 ranked configurations on 25 Oct 2026. 443 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 4 31B cost per 1M tokens?
- As of 25 Oct 2026, Gemma 4 31B costs $0.200 per 1M input tokens and $0.400 per 1M output tokens, $0.250 blended (three input tokens per output token). The list price last changed on 21 Oct 2026.
- When was Gemma 4 31B released?
- Google released Gemma 4 31B on 2 Apr 2026. BenchLeader has results for it at 5 reasoning-effort settings.
Data as of 25 Oct 2026. Compare these configurations.
Cite as: BenchLeader, “Gemma 4 31B: benchmarks, pricing, speed and rank”, data as of 25 Oct 2026, https://www.benchleader.com/models/gemma-4-31b.