modelgrep
G

Google: Gemma 4 31B

google/gemma-4-31b-it

48th smartest of 178Cheaper than 78% of paidReasoningToolsJSONVision
Use via OpenRouter ↗
Intelligence
39.2
48th of 178
Design Elo
Speed
55
150th fastest
Latency
309ms
first token
Input price
$0.120
66th cheapest
Context
262K
262K max out

How it compares

Smarter than73%
of all ranked models
Faster than49%
of all ranked models
Cheaper than78%
of all ranked models

Overview

Gemma 4 31B Instruct is Google DeepMind's 30.7B dense multimodal model supporting text and image input with text output. Features a 256K token context window, configurable thinking/reasoning mode, native function...

Benchmarks

independent · via OpenRouter
Artificial Analysis71th percentile
Intelligence Index
39.2
Coding Index
38.7
Agentic Index
40.9
GPQA Diamond
86%
Humanity's Last Exam
23%
SciCode
43%
Tau²-Bench (agentic)
60%

Providers & pricing (11)

ProviderIn $/MOut $/MUptime
WandBbf16$0.120$0.35099.9%
Venicebf16$0.120$0.36098%
DeepInfrafp4$0.120$0.37097.5%
DeepInfrafp8$0.130$0.38097.4%
SiliconFlowfp8$0.130$0.40088%
Novitabf16$0.140$0.40099.4%
Parasailfp8$0.150$0.40093.9%
Chutesfp4$0.150$0.42097.3%
Phala$0.150$0.46090.9%
Together$0.280$0.86092.5%
Together$0.390$0.97096.7%

Specifications

Context window262K
Max output262K
Knowledge cutoff
Input modalitiesimage, text, video
Output modalitiestext
Prompt caching
Cache read price$0.090/M
ModeratedNo

Gemma 4 31B FAQ

How much does Gemma 4 31B cost?

Gemma 4 31B costs $0.120 per million input tokens and $0.350 per million output tokens via OpenRouter, making it 66th cheapest of 298 paid models.

How smart is Gemma 4 31B?

Gemma 4 31B scores 39.2 on the Artificial Analysis Intelligence Index, ranking 48th of 178 benchmarked models, with a GPQA Diamond score of 86%.

How fast is Gemma 4 31B?

Gemma 4 31B generates around 55 tokens per second with 309ms time-to-first-token (p50), the 150th fastest tracked model.

What is Gemma 4 31B's context window?

Gemma 4 31B supports a 262K-token context window and can output up to 262K tokens. It accepts image, text, video input.

Compare head-to-head