Gemma 4 31B vs Gemma 4 26B A4B
Gemma 4 31B scores higher on 8 of the 8 benchmarks both models publish. The widest gap is on Humanity's Last Exam, where Gemma 4 31B leads by 10.8 points.
Gemma 4 31BGoogle · OpenLLM Buddy catalog79.2compositeGemma 4 26B A4BGoogle · OpenLLM Buddy catalog72.9compositeShared benchmarks
Raw published scores; the higher score on each row is highlighted. LMArena Elo is a human-preference vote and is not counted as a win.
- GPQA Diamond84.382.3
- Humanity's Last Exam19.58.7
no toolsno tools
- Humanity's Last Exam (with tools)26.517.2
with searchwith search
- AIME89.288.3
AIME 2026, no toolsAIME 2026, no tools
- LiveCodeBench80.077.1
LiveCodeBench v6LiveCodeBench v6
- τ²-bench76.968.2
avg. over 3 domainsavg. over 3 domains
- MMLU-Pro85.282.6
- MMMU-Pro76.973.8
MMMU-ProMMMU-Pro
- LMArena Text Elo · not counted14521438
LMArena text overall, 'Gemma 4 31B', ±7, as of 2026-10-27LMArena text overall, 'Gemma 4 26B A4B', ±7, as of 2026-10-27
Should you pick Gemma 4 31B or Gemma 4 26B A4B?
Pick Gemma 4 31B if…
- You care most about GPQA Diamond, Humanity's Last Exam, Humanity's Last Exam (with tools), where it scores at least 2 points higher.
- People prefer its answers in blind chats: LMArena Elo 1452 against 1438.
- Cost matters: $0.340 per million output tokens against $0.600.
- You want it running in minutes on a dedicated GPU through OpenLLM Buddy, in an uncensored build.
Pick Gemma 4 26B A4B if…
- You want the lighter model to host: about 17 GB at 4-bit against 21 GB.
- You need fast responses: about 113 tokens per second through APIs, against 35.
- You want it running in minutes on a dedicated GPU through OpenLLM Buddy.
Hardware and cost to run
Gemma 4 31B
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)).
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.
Deploy Gemma 4 31BGemma 4 26B A4B
At 4-bit it needs about 17 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 58 GB (1× H100 (80 GB)).
On OpenLLM Buddy: Gemma 4 26B A4B, with an OpenAI-compatible API and no per-token fees. Prices start at $13 USD for 11 Hours.
Deploy Gemma 4 26B A4BCategory profile
Category index uses every test each model reports, not only the shared ones.
Specs side by side
| Gemma 4 31B | Gemma 4 26B A4B | |
|---|---|---|
| Lab | ||
| Weights | Open | Open |
| Released | 2026-04-02 | 2026-04-02 |
| Parameters | 30.7B dense | 25.2B total, 3.8B active (MoE) |
| Context | 256K | 256K |
| License | Apache-2.0 | Apache-2.0 |
| 4-bit memory | 21 GB · RTX 4090 (24 GB) | 17 GB · RTX 4090 (24 GB) |
| API price | $0.100 in / $0.340 out per 1M tokens | $0.150 in / $0.600 out per 1M tokens |
| API speed | 35 tok/s | 113 tok/s |
| One-GPU speed (est.) | ≈ 25–34 tok/s on RTX 4090 (4-bit) | ≈ 133–160 tok/s on RTX 4090 (4-bit) |
Related comparisons
Sources
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