Google · Benchmarks

Gemma 4 31B benchmarks

Gemma 4 31B has 9 published scores in this dataset, from a Google release dated 2026-04-02. Its composite index is 79.2, which places it 16th of 22 ranked models and 11th among 15 open-weight models. Its strongest area is knowledge (index 92.5) and its weakest is agentic (index 68.9). Its best placing is on τ²-bench, where its 76.9 ranks 1st of 2 models with a published score.

Deploy hereDeploy gemma-4-31B-it-uncensoredOfficial page

OpenLLM Buddy runs gemma-4-31B-it-uncensored, an uncensored fine-tune of Gemma 4 31B, on RTX 4090 24GB (24 GB VRAM). The scores below are the official base model's; a fine-tune can score slightly differently.

Category profile

Index 0–100: each score relative to the best published score on that test.

ReasoningCodingAgenticKnowledge
Reasoning
69.8#13 of 16 · 3 tests
Coding
85.61 test · not ranked
Agentic
68.9#6 of 8 · 2 tests
Knowledge
92.5#3 of 4 · 2 tests

Model facts

Lab
Google
Type
OpenLLM Buddy catalog
Released
2026-04-02
Parameters
30.7B dense
Context
256K
License
Apache-2.0
Runs on
RTX 4090 24GB

Gemma 4 31B hardware requirements and cost

At 4-bit it needs about 21 GB, which fits a single RTX 4090, so it can run on a home workstation with llama.cpp, Ollama, or vLLM. Full BF16 precision needs about 71 GB (1× H100 (80 GB)).

Estimated GPU memory for Gemma 4 31B
PrecisionWeightsWith headroomSmallest setup
BF16 (full precision)61 GB71 GB1× H100 (80 GB)
FP8 / 8-bit31 GB35 GB1× H100 (80 GB)
4-bit (GGUF Q4_K_M)18 GB21 GBRTX 4090 (24 GB)

Estimate: parameters × bytes per parameter, plus 15% for the KV cache and runtime at modest context lengths. Long contexts and large batches need more.

On OpenLLM Buddy: gemma-4-31B-it-uncensored runs on RTX 4090 24GB, with an OpenAI-compatible API and no per-token fees. Prices start at $13 USD for 11 hours.

  • 11 hours: $13 USD
  • 24 hours: $27 USD
  • 1 week: $184 USD
  • 1 month: $736 USD
Deploy Gemma 4 31B

Lighter or stronger open models

Open models ranked just above and below on the composite index, with their 4-bit memory needs.

Gemma 4 31B speed and pricing

Measured figures link to their source; estimates are labelled.

API output speed
35 tok/sArtificial Analysis median · Artificial Analysis
Time to first token
0.76 sOpenRouter model page provider stats
API price per 1M tokens
$0.100 in · $0.340 outcheapest provider on OpenRouter: CoreWeave · OpenRouter endpoints API
Speed on OpenLLM Buddy (RTX 4090, 4-bit)estimate
25–34 tok/s
OpenLLM Buddy GPU rental
from $1.10 per hourflat rate, no per-token fees
Cost per 1M output tokens, one userestimate
≈ $10.20falls sharply when several requests share the GPU
On your own RTX 4090 (4-bit)estimate
25–34 tok/s

Estimate for one user on one GPU: active weights read per token ÷ memory bandwidth at 50–70% efficiency, plus 3 ms overhead per token. Real speed varies with engine, context length and quantization; serving several users at once raises total throughput.

Compare speed and price across all models

Score by test

Rank is among every model with a published score on that test.

GPQA Diamond84.3%Graduate-level, Google-proof science questions in biology, physics, and chemistry.#17 of 19 · 11.8 behind GPT-6 AstraHumanity's Last Exam19.5%no tools#21 of 23 · 44.9 behind Claude Opus 5.5Humanity's Last Exam (with tools)26.5%with search#15 of 17 · 43.5 behind Kimi K3.1 ProAIME89.2%AIME 2026, no tools#4 of 5 · 8.5 behind Reflection Beam 1.0LiveCodeBench80.0%LiveCodeBench v6#6 of 7 · 13.5 behind DeepSeek V4 Proτ²-bench76.9%avg. over 3 domains#1 of 2 · top scoreMMLU-Pro85.2%Broad academic knowledge with ten answer options instead of four.#4 of 5 · 2.3 behind DeepSeek V4 ProMMMU-Pro76.9%MMMU-Pro#7 of 9 · 10.8 behind Claude Opus 5.5LMArena Text Elo1452LMArena text overall, 'Gemma 4 31B', ±7, as of 2026-10-27#14 of 21 · 68 behind Gemini 4 Argon

Nearby in the composite ranking

Models with the closest composite index.

  1. 1Qwen3.8 27B83.2
  2. 2GLM 5.382.7
  3. 3DeepSeek V4.1 Flash79.8
  4. 4Gemma 4 31B79.2
  5. 5Step 5 Preview78.7
  6. 6Nemotron 3 Ultra74.4
  7. 7Gemma 4 26B A4B72.9

Head-to-head comparisons

Side-by-side pages built from the tests both models report.

Sources

ModelTestScoreSettingSource
Gemma 4 31BGPQA Diamond84.3—Gemma 4 31B IT model card (Hugging Face)
Gemma 4 31BHumanity's Last Exam19.5no toolsGemma 4 31B IT model card (Hugging Face)
Gemma 4 31BHumanity's Last Exam (with tools)26.5with searchGemma 4 31B IT model card (Hugging Face)
Gemma 4 31BAIME89.2AIME 2026, no toolsGemma 4 31B IT model card (Hugging Face)
Gemma 4 31BLiveCodeBench80.0LiveCodeBench v6Gemma 4 31B IT model card (Hugging Face)
Gemma 4 31Bτ²-bench76.9avg. over 3 domainsGemma 4 31B IT model card (Hugging Face)
Gemma 4 31BMMLU-Pro85.2—Gemma 4 31B IT model card (Hugging Face)
Gemma 4 31BMMMU-Pro76.9MMMU-ProGemma 4 31B IT model card (Hugging Face)
Gemma 4 31BLMArena Text Elo1452LMArena text overall, 'Gemma 4 31B', ±7, as of 2026-10-27Arena (LMArena) text leaderboard

Scores are published results from model cards, lab reports, and public leaderboards, each linked to its source. Settings such as thinking mode or test version differ between sources and are shown next to each score. OpenLLM Buddy catalog models are uncensored fine-tunes; their scores are those of the official base model. Data collected 2026-10-30. How the index works