Gemma 4 12B Opus Reasoning
yuxinlu1/gemma-4-12b-opus-reasoning
yuxinlu1/gemma-4-12B-it-Claude-4.6-4.8-Opus-GGUF on HuggingFace ↗
Throughput
- @ 4k 19.1t/s
Decode tok/s after a fixed context fill (PBM) — same agent-style workload as bench v2.
Recipe
- Profile
- gemma4-12b-opus-reasoning-q4
- Engine
- llama.cpp
- Context
- 32k · q8_0 KV
- Served as
- gemma4-12b-opus-reasoning-q4
Run on your Spark
single node
shell
spark inference up gemma4-12b-opus-reasoning-q4
Why we run it
Golden fleet target — auto-scaffolded from recipe gemma4-12b-opus-reasoning-q4.
Bench notes
PBM 4k @ 19.1 tok/s — perfbench-metrics — profile=gemma4-12b-opus-reasoning-q4
Measurement history
Context ladder
Older bench-v2 / golden cells at each benched context window (single measurement).
| Context | KV | Throughput |
|---|---|---|
| @ 32k peak golden | — | 17.3t/s |
Benchmark runs
Recorded inference benchmark sessions for this model's profile (single run).
| Date | Profile | Method | Avg | Session t/s | Range | Fill | Tool |
|---|---|---|---|---|---|---|---|
| 2026-06-28latest | gemma4-12b-opus-reasoning-q4 | bench-agent-v2v2.0 | 17.3t/s | — | — | ~14,745 | ok |
| golden 32k/q8_0 @ 17.3 tok/s — fill~14745 — bench-agent-v2 — tool_ok=True | |||||||