Head to head

Qwen3.8 27B vs Gemma 4 31B

Qwen3.8 27B scores higher on 3 of the 3 benchmarks both models publish. The widest gap is on Humanity's Last Exam, where Qwen3.8 27B leads by 11.3 points.

Qwen3.8 27BAlibaba Qwen · OpenLLM Buddy catalog83.2compositeGemma 4 31BGoogle · OpenLLM Buddy catalog79.2composite

Shared 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.

Should you pick Qwen3.8 27B or Gemma 4 31B?

Pick Qwen3.8 27B if…

  • You care most about GPQA Diamond, Humanity's Last Exam, LiveCodeBench, where it scores at least 2 points higher.
  • You want the lighter model to host: about 19 GB at 4-bit against 21 GB.
  • You want it running in minutes on a dedicated GPU through OpenLLM Buddy, in an uncensored build.

Pick Gemma 4 31B if…

  • People prefer its answers in blind chats: LMArena Elo 1452 against 1436.
  • Cost matters: $0.340 per million output tokens against $2.55.
  • You want it running in minutes on a dedicated GPU through OpenLLM Buddy, in an uncensored build.

Hardware and cost to run

Qwen3.8 27B

At 4-bit it needs about 19 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 62 GB (1× H100 (80 GB)).

On OpenLLM Buddy: Qwen3.8-27B-Uncensored-FP8 runs on H100 80GB as an FP8 build, with an OpenAI-compatible API and no per-token fees. Prices start at $53 USD for 11 hours.

Deploy Qwen3.8 27B
Full GPU memory table for Qwen3.8 27B

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 31B
Full GPU memory table for Gemma 4 31B

Category profile

Category index uses every test each model reports, not only the shared ones.

ReasoningCodingAgenticKnowledge
Qwen3.8 27BGemma 4 31B
Reasoning70.369.8
Coding81.985.6
Agentic—68.9
Knowledge97.392.5

Specs side by side

Qwen3.8 27BGemma 4 31B
LabAlibaba QwenGoogle
WeightsOpenOpen
Released2026-08-142026-04-02
Parameters27B dense30.7B dense
Context262K (up to 1M)256K
LicenseApache-2.0Apache-2.0
4-bit memory19 GB · RTX 4090 (24 GB)21 GB · RTX 4090 (24 GB)
API price$0.425 in / $2.55 out per 1M tokens$0.100 in / $0.340 out per 1M tokens
API speed41 tok/s35 tok/s
One-GPU speed (est.)≈ 52–69 tok/s on H100 80GB (FP8)≈ 25–34 tok/s on RTX 4090 (4-bit)

Related comparisons

Sources

ModelTestScoreSettingSource
Qwen3.8 27BGPQA Diamond89.2—Qwen3.8-27B model card (Hugging Face)
Gemma 4 31BGPQA Diamond84.3—Gemma 4 31B IT model card (Hugging Face)
Qwen3.8 27BHumanity's Last Exam30.8—Qwen3.8-27B model card (Hugging Face)
Gemma 4 31BHumanity's Last Exam19.5no toolsGemma 4 31B IT model card (Hugging Face)
Qwen3.8 27BLiveCodeBench90.3LiveCodeBench v6Qwen3.8-27B model card (Hugging Face)
Gemma 4 31BLiveCodeBench80.0LiveCodeBench v6Gemma 4 31B IT model card (Hugging Face)
Qwen3.8 27BLMArena Text Elo1436LMArena text overall, 'Qwen3.8-27B', ±5, as of 2026-10-27Arena (LMArena) text leaderboard
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