vs
0:14
overall capability wins (1 ties across 15 scored). Gemma 4 31B leads overall.
- Reasoning0:7
- Coding0:3
- Arena0:1
- Tool use0:1
- Composite indices0:2
- Pricing3:0
- Speed2:0
- Specs0:2
Verdict
Gemma 4 31B leads coding (widest gap: +55.4 on LiveCode); Gemma 4 31B ranks higher on LMArena (+109 Elo); Gemma 4 31B offers ~2.0x larger context (256K vs 128K).
- Gemma 4 31B leads coding (widest gap: +55.4 on LiveCode)
- Gemma 4 31B ranks higher on LMArena (+109 Elo)
- Gemma 4 31B offers ~2.0x larger context (256K vs 128K)
- Gemma 3 12B is ~2.0x cheaper on a blended token basis than Gemma 4 31B
Cost for 1M in + 250K out
Illustrative chat workload at primary-provider list prices.
- Gemma 3 12B
- $0.09
- Gemma 4 31B
- $0.17
Gemma 3 12B is about 2.0x cheaper on this workload.
Price per 1M tokens
Full comparison
Organization
Family
Gemma
Gemma
License
Gemma
Apache 2.0
Open weights
Yes
Yes
Release
Mar 12, 2025
Apr 8, 2026
Knowledge cutoff
-
-
API / provider
Google AI / self-host
Google AI / self-host
Modalities
text, image → text
text, image → text
Specs
Context window
128K
256K
Max output
8K
8K
Parameters
12B
31B
Pricing
Input $/1M
$0.05
$0.10
Output $/1M
$0.15
$0.30
Blended $/1M (3∶1)
$0.08
$0.15
Speed
tok/s
120
85
TTFT (s)
0.22
0.3
Reasoning
MMLU-Pro
60.6
85.2
GPQA Diamond
34.9
85.7
Humanity's Last Exam
4.2
23.6
AIME 2025
18.3
—
MATH-500
85.3
—
Humanity's Last Exam (with tools)
—
—
AA-Omniscience Accuracy
10.3
20.0
AA-LCR v1.1
8.3
69.7
CritPt
—
1.4
MMMU-Pro
37.5
73.4
IFBench
36.7
75.6
Chartography
—
—
Chartography (With Tools)
—
—
Coding
SWE-bench Verified
—
—
SWE-bench Pro
—
—
SWE-bench Multilingual
—
—
LiveCodeBench
24.6
80.0
Terminal-Bench 2.1
0.0
43.4
Aider Polyglot
—
—
Terminal-Bench 3
—
—
BigCodeBench
—
—
SciCode
16.4
45.5
CursorBench
—
—
SWE-Rebench
—
25.1
NL2Repo-Bench
—
—
DeepSWE
—
—
WebDev Arena
—
1,364
Terminal-Bench 4.0
—
—
CursorBench 4.0
—
—
FrontierCode v1.1 (Main)
—
—
Arena
LMArena Elo
1,342
1,451
| Metric | Gemma 3 12B | Gemma 4 31B | Delta |
|---|---|---|---|
| Identity | |||
| Organization | — | ||
| Family | Gemma | Gemma | — |
| License | Gemma | Apache 2.0 | — |
| Open weights | Yes | Yes | — |
| Release | Mar 12, 2025 | Apr 8, 2026 | — |
| Knowledge cutoff | - | - | — |
| API / provider | Google AI / self-host | Google AI / self-host | — |
| Modalities | text, image → text | text, image → text | — |
| Specs | |||
| Context window | 128K | 256K | B +128K |
| Max output | 8K | 8K | tie |
| Parameters | 12B | 31B | B +19 |
| Pricing | |||
| Input $/1M | $0.05 | $0.10 | A +$0.05 |
| Output $/1M | $0.15 | $0.30 | A +$0.15 |
| Blended $/1M (3∶1) | $0.08 | $0.15 | A +$0.08 |
| Speed | |||
| tok/s | 120 | 85 | A +35 |
| TTFT (s) | 0.22 | 0.3 | A +0.08 |
| Reasoning | |||
| MMLU-Pro | 60.6 | 85.2 | B +24.6 pts |
| GPQA Diamond | 34.9 | 85.7 | B +50.8 pts |
| Humanity's Last Exam | 4.2 | 23.6 | B +19.4 pts |
| AIME 2025 | 18.3 | — | — |
| MATH-500 | 85.3 | — | — |
| Humanity's Last Exam (with tools) | — | — | — |
| AA-Omniscience Accuracy | 10.3 | 20.0 | B +9.7 pts |
| AA-LCR v1.1 | 8.3 | 69.7 | B +61.4 pts |
| CritPt | — | 1.4 | — |
| MMMU-Pro | 37.5 | 73.4 | B +35.9 pts |
| IFBench | 36.7 | 75.6 | B +38.9 pts |
| Chartography | — | — | — |
| Chartography (With Tools) | — | — | — |
| Coding | |||
| SWE-bench Verified | — | — | — |
| SWE-bench Pro | — | — | — |
| SWE-bench Multilingual | — | — | — |
| LiveCodeBench | 24.6 | 80.0 | B +55.4 pts |
| Terminal-Bench 2.1 | 0.0 | 43.4 | B +43.4 pts |
| Aider Polyglot | — | — | — |
| Terminal-Bench 3 | — | — | — |
| BigCodeBench | — | — | — |
| SciCode | 16.4 | 45.5 | B +29.1 pts |
| CursorBench | — | — | — |
| SWE-Rebench | — | 25.1 | — |
| NL2Repo-Bench | — | — | — |
| DeepSWE | — | — | — |
| WebDev Arena | — | 1,364 | — |
| Terminal-Bench 4.0 | — | — | — |
| CursorBench 4.0 | — | — | — |
| FrontierCode v1.1 (Main) | — | — | — |
| Arena | |||
| LMArena Elo | 1,342 | 1,451 | B +109 Elo |
Gemma 3 12B
- Context
- 128K / 8K out
- Parameters
- 12B
- Price
- $0.05 / $0.15
- Speed
- 120 tok/s · 0.22s TTFT
- Modalities
- text, image → text
- License
- Gemma
Compact open Gemma 3 instruct model for edge and low-cost multimodal chat.
Gemma 4 31B
- Context
- 256K / 8K out
- Parameters
- 31B
- Price
- $0.10 / $0.30
- Speed
- 85 tok/s · 0.3s TTFT
- Modalities
- text, image → text
- License
- Apache 2.0
Open-weight Gemma 4 instruction model - strong mid-size open alternative under Apache 2.0.
Benchmark charts
Winner bars are emphasized. Per-benchmark deltas sit above each chart.