Gemma 4 12B vs Step 3.7 Flash: Benchmarks, Pricing & Capabilities Compared
TL;DR — Gemma 4 12B wins for general use · Step 3.7 Flash wins for reasoning + long-context.
Gemma 4 12B Google
- Released
- 2026-06-03
- Context window
- 131K tokens
- Input price
- $0.00 / Mtok
- Output price
- $0.00 / Mtok
Step 3.7 Flash StepFun
- Released
- 2026-05-29
- Context window
- 256K tokens
- Input price
- $0.20 / Mtok
- Output price
- $1.15 / Mtok
Benchmark comparison
| Benchmark | Gemma 4 12B | Step 3.7 Flash |
|---|---|---|
| GPQA Diamond | 75.3% | 80.9% ✓ |
| HLE | 14.8% | 19.9% ✓ |
| IF-Bench | 73.5% ✓ | 67.3% |
| LiveCodeBench Reasoning | 55.3% | 63.7% ✓ |
| SciCode | 38.2% | 40.0% ✓ |
| TAU2-bench | 36.3% | 98.5% ✓ |
| TerminalBench-Hard | 18.2% | 35.6% ✓ |
Pricing comparison
| Metric | Gemma 4 12B | Step 3.7 Flash |
|---|---|---|
| Input ($/Mtok) | $0.00 | $0.20 |
| Output ($/Mtok) | $0.00 | $1.15 |
| Cached input ($/Mtok) | — | — |
| Cost per 1M-token roundtrip (1M in + 1M out) | $0.00 | $1.35 |
Context window & modalities
| Attribute | Gemma 4 12B | Step 3.7 Flash |
|---|---|---|
| Context window | 131K tokens | 256K tokens |
| Input modalities | text, image, video | text, image |
| Output modalities | text | text |
| Knowledge cutoff | — | — |
Verdict by use case
Coding
Insufficient data
Basis: SWE-bench
No shared coding benchmark.
Reasoning
→ Step 3.7 Flash
Basis: GPQA Diamond
Gemma 4 12B 75.3% vs Step 3.7 Flash 80.9% on GPQA Diamond.
Math
Insufficient data
Basis: MATH / AIME
No shared math benchmark.
Long context
→ Step 3.7 Flash
Basis: Context window
Gemma 4 12B 131K tokens vs Step 3.7 Flash 256K tokens.
Cost
→ Gemma 4 12B
Basis: Input $/Mtok
Gemma 4 12B $0/Mtok vs Step 3.7 Flash $0.2/Mtok input.
Changelog & releases
Gemma 4 12B
Released 2026-06-03
Step 3.7 Flash
Released 2026-05-29