How accurate is Gemma 4 31B?

Gemma 4 31B answered 24 of 80 questions correctly (30.0%) in our 2026-10-27 run, ranking #64 of 78 models.

How fast is Gemma 4 31B?

Gemma 4 31B generated a median of 58 output tokens per second in our 2026-10-27 run, ranking #54 of 78. Median time to first token was 0.6 s.

What is the latency of Gemma 4 31B?

Gemma 4 31B had a median time to first token of 0.6 s in our 2026-10-27 run, ranking #26 of 78 models.

How much does Gemma 4 31B cost?

Gemma 4 31B cost $0.0014 per question on average as billed in our 2026-10-27 run, reasoning tokens included.

What is Gemma 4 31B API pricing?

Gemma 4 31B is listed at $0.2 per million input tokens and $0.4 per million output tokens on OpenRouter.

What is the context window of Gemma 4 31B?

Gemma 4 31B accepts up to 262,144 tokens of context, according to OpenRouter's model catalogue.

Is Gemma 4 31B open source?

Yes. Gemma 4 31B is an open-weights model: Google publishes the weights, so you can run it yourself as well as through an API.

Who makes Gemma 4 31B?

Gemma 4 31B is made by Google. We tested it through OpenRouter using the model ID google/gemma-4-31b-it.

Where can I use Gemma 4 31B via API?

Through OpenRouter, Gemma 4 31B is served by 3 hosts: CoreWeave, DeepInfra, ModelRun.

GoogleOpen weightsReleased 2 Apr 2026

Gemma 4 31B benchmarks, speed and API pricing

Reviewed by LLM Verifications team · Tested 27 Oct 2026

Compare

Available through OpenRouter on

DeepInfra$0.16 in · $0.32 out
CoreWeave$0.1 in · $0.34 out
Venice$0.12 in · $0.36 out
Chutes$0.12 in · $0.37 out
DekaLLM$0.14 in · $0.4 out
Crusoe$0.14 in · $0.4 out
Friendli$0.14 in · $0.4 out
Novita$0.14 in · $0.4 out
Parasail$0.15 in · $0.4 out
ModelRun$0.75 in · $1 out
SiliconFlow$0.75 in · $1 out
Io Net$0.361 in · $1.0925 out
SambaNova$0.38 in · $1.15 out

Intelligence30.0% correct#64 of 72

Speed58tok/s#54 of 72

Cost$0.0014per question#32 of 72List $0.2 in · $0.4 out per 1M

Verbosity3,307tokens per question#67 of 72

Summary

Answers 24 of 80 verification questions correctly (30.0%), placing #64 of 72 models tested on the current suite.
Generates 58 output tokens per second, faster than 25% of the models we measure.
Starts responding after 0.6 s (median time to first token), including thinking time.
Costs $0.0014 per question as billed, #18 cheapest of 29 models priced $0.15 to $1 per 1M tokens.

Specifications

Reasoning: Yes
Input: image, text, video
Output: text
Context window: 262k tokens
Weights: Open weights
Knowledge cutoff: —
OpenRouter ID: google/gemma-4-31b-it
Last tested: 27 Oct 2026

Peer group: 31 models with a blended list price $0.15 to $1 per 1M tokens (3 input : 1 output). Charts below compare Gemma 4 31B with all models; cost and speed charts focus on this peer group.

Intelligence

Verification score: the share of 80 questions from open datasets answered correctly. See the benchmark datasets.

Score vs. cost per question

Gemma 4 31B circled; dashed line joins models no cheaper model beats

Score vs. end-to-end response time

Gemma 4 31B circled · accurate and quick sits top left

Score vs. tokens per question

Gemma 4 31B circled · output tokens, reasoning included

Verification score

All models tested on the current suite

61 more models · see the full leaderboard

Cost

Average cost of one question, split into input, reasoning and answer tokens at each model's list price. Reasoning tokens are billed as output.

Cost per question, by token type

Peer group: blended price $0.15 to $1

1Llama 4 Scout$0.00025
2GPT-6 Luna$0.00025
3Ministral 3 8B 2512$0.00027
4Llama 3.3 70B Instruct$0.00031
5Ministral 3 14B 2512$0.00031
6GLM 5.3 Flash$0.00033
7gpt-oss-120b$0.00038
8Inkling Small$0.00047
9Claude Haiku 5.5$0.00048
10MiMo-V2.6-Flash$0.00056
11Llama 4 Maverick$0.00057
12GLM 5.3$0.00076
13DeepSeek V4.1 Flash 1017$0.00084
14MiniMax M3.1 Flash$0.0009
15DeepSeek V4 Flash Vision Exp$0.00092
16Hy3$0.0013
17Gemma 4 31B$0.0014
18Nemotron 3.5 Super VL$0.0014
19Qwen3 Coder Next$0.0017
20Solar Pro 4$0.0018
21Mistral Small 4$0.0019
22MiMo-V2.6-Pro$0.002
23Trinity Large Thinking$0.0025
24MiniMax M3$0.0027
25LongCat 2.0$0.0032
26Qwen3.7 Plus$0.0033
27Gemini 3.5 Flash Lite$0.0034
28Qwen3.6 35B A3B$0.0034
29Qwen3.8 27B$0.0049

Input Reasoning Answer

Cost per question, as billed

All models · lower is better

1Granite 4.0 Micro$0.000051
2gpt-oss-20b$0.000066
3Granite 4.2 8B$0.000077
4Nova Micro 1.0$0.000087
5Mercury 2.5$0.00011
6Phi 4$0.00012
7gpt-oss-120b$0.00013
8GLM 5.3 Flash$0.00015
9Ministral 3 3B 2512$0.00019
10GPT-6 Luna$0.00025
32Gemma 4 31B$0.0014

61 more models · see the full leaderboard

Speed and latency

Median of 3 streamed runs of the same prompt, measured 25 Oct 2026.

Output speed

Peer group ($0.15 to $1) · higher is better

16 more models · see the full leaderboard

Time to first token

Peer group ($0.15 to $1) · lower is better

16 more models · see the full leaderboard

End-to-end response time

Peer group ($0.15 to $1) · lower is better

16 more models · see the full leaderboard

Latency vs. output speed

Gemma 4 31B circled · quick to start and fast sits top left

Results by category

Questions answered correctly in each category

Science11/30

Engineering0/15

Medicine2/11

Economics5/9

Law1/6

Humanities & social science5/9

More from Google

Google#2Gemini 4 Argon84.8% verification score69 tok/s3.3 s first token$0.0063 / question

Google#3Gemini 3.8 Flash82.5% verification score142 tok/s3.2 s first token$0.0037 / question

Google#9Gemini 3.1 Pro Preview77.5% verification score124 tok/s3.4 s first token$0.017 / question

All Google models →

Compare Gemma 4 31B

Head to head with its closest rivals, other Google models and the top models overall.

Hosts and prices

Every host serving this model through OpenRouter, from OpenRouter's catalogue

Host	Input / 1M	Output / 1M	Context	Max output	Quantization
DeepInfra	$0.16	$0.32	262k	236k	fp8
CoreWeave	$0.1	$0.34	262k	236k	fp4
Venice	$0.12	$0.36	256k	8k	fp4
Chutes	$0.12	$0.37	131k	66k	fp4
DekaLLM	$0.14	$0.4	262k	236k	—
Crusoe	$0.14	$0.4	262k	262k	bf16
Friendli	$0.14	$0.4	262k	8k	—
Novita	$0.14	$0.4	262k	131k	bf16
Parasail	$0.15	$0.4	262k	236k	fp8
ModelRun	$0.75	$1	262k	236k	fp4
SiliconFlow	$0.75	$1	262k	236k	fp8
Io Net	$0.361	$1.0925	262k	16k	—
SambaNova	$0.38	$1.15	131k	118k	—

Sources

Scores, speed and billed cost are our own measurements. Prices, context window and hosts come from OpenRouter's catalogue on the test date.

Gemma 4 31B on OpenRouter: list prices and hosts
Model card on Hugging Face: weights and licence
SuperGPQA (paper): 50 of the questions, ODC-BY 1.0
MMLU-Pro (paper): 30 of the questions, MIT
OpenRouter generation records: where billed cost and native token counts come from
Our methodology and editorial standards

Gemma 4 31B: frequently asked questions

How accurate is Gemma 4 31B?: Gemma 4 31B answered 24 of 80 questions correctly (30.0%) in our 2026-10-27 run, ranking #64 of 78 models.
How fast is Gemma 4 31B?: Gemma 4 31B generated a median of 58 output tokens per second in our 2026-10-27 run, ranking #54 of 78. Median time to first token was 0.6 s.
What is the latency of Gemma 4 31B?: Gemma 4 31B had a median time to first token of 0.6 s in our 2026-10-27 run, ranking #26 of 78 models.
How much does Gemma 4 31B cost?: Gemma 4 31B cost $0.0014 per question on average as billed in our 2026-10-27 run, reasoning tokens included.
What is Gemma 4 31B API pricing?: Gemma 4 31B is listed at $0.2 per million input tokens and $0.4 per million output tokens on OpenRouter.
What is the context window of Gemma 4 31B?: Gemma 4 31B accepts up to 262,144 tokens of context, according to OpenRouter's model catalogue.
Is Gemma 4 31B open source?: Yes. Gemma 4 31B is an open-weights model: Google publishes the weights, so you can run it yourself as well as through an API.
Who makes Gemma 4 31B?: Gemma 4 31B is made by Google. We tested it through OpenRouter using the model ID google/gemma-4-31b-it.
Where can I use Gemma 4 31B via API?: Through OpenRouter, Gemma 4 31B is served by 3 hosts: CoreWeave, DeepInfra, ModelRun.

GoogleOpen weightsReleased 2 Apr 2026

Gemma 4 31B benchmarks, speed and API pricing

Reviewed by LLM Verifications team · Tested 27 Oct 2026

Compare

Available through OpenRouter on

DeepInfra$0.16 in · $0.32 out
CoreWeave$0.1 in · $0.34 out
Venice$0.12 in · $0.36 out
Chutes$0.12 in · $0.37 out
DekaLLM$0.14 in · $0.4 out
Crusoe$0.14 in · $0.4 out
Friendli$0.14 in · $0.4 out
Novita$0.14 in · $0.4 out
Parasail$0.15 in · $0.4 out
ModelRun$0.75 in · $1 out
SiliconFlow$0.75 in · $1 out
Io Net$0.361 in · $1.0925 out
SambaNova$0.38 in · $1.15 out

Intelligence30.0% correct#64 of 72

Speed58tok/s#54 of 72

Cost$0.0014per question#32 of 72List $0.2 in · $0.4 out per 1M

Verbosity3,307tokens per question#67 of 72

Summary

Answers 24 of 80 verification questions correctly (30.0%), placing #64 of 72 models tested on the current suite.
Generates 58 output tokens per second, faster than 25% of the models we measure.
Starts responding after 0.6 s (median time to first token), including thinking time.
Costs $0.0014 per question as billed, #18 cheapest of 29 models priced $0.15 to $1 per 1M tokens.

Specifications

Reasoning: Yes
Input: image, text, video
Output: text
Context window: 262k tokens
Weights: Open weights
Knowledge cutoff: —
OpenRouter ID: google/gemma-4-31b-it
Last tested: 27 Oct 2026

Peer group: 31 models with a blended list price $0.15 to $1 per 1M tokens (3 input : 1 output). Charts below compare Gemma 4 31B with all models; cost and speed charts focus on this peer group.

Intelligence

Verification score: the share of 80 questions from open datasets answered correctly. See the benchmark datasets.

Score vs. cost per question

Gemma 4 31B circled; dashed line joins models no cheaper model beats

Score vs. end-to-end response time

Gemma 4 31B circled · accurate and quick sits top left

Score vs. tokens per question

Gemma 4 31B circled · output tokens, reasoning included

Verification score

All models tested on the current suite

61 more models · see the full leaderboard

Cost

Average cost of one question, split into input, reasoning and answer tokens at each model's list price. Reasoning tokens are billed as output.

Cost per question, by token type

Peer group: blended price $0.15 to $1

1Llama 4 Scout$0.00025
2GPT-6 Luna$0.00025
3Ministral 3 8B 2512$0.00027
4Llama 3.3 70B Instruct$0.00031
5Ministral 3 14B 2512$0.00031
6GLM 5.3 Flash$0.00033
7gpt-oss-120b$0.00038
8Inkling Small$0.00047
9Claude Haiku 5.5$0.00048
10MiMo-V2.6-Flash$0.00056
11Llama 4 Maverick$0.00057
12GLM 5.3$0.00076
13DeepSeek V4.1 Flash 1017$0.00084
14MiniMax M3.1 Flash$0.0009
15DeepSeek V4 Flash Vision Exp$0.00092
16Hy3$0.0013
17Gemma 4 31B$0.0014
18Nemotron 3.5 Super VL$0.0014
19Qwen3 Coder Next$0.0017
20Solar Pro 4$0.0018
21Mistral Small 4$0.0019
22MiMo-V2.6-Pro$0.002
23Trinity Large Thinking$0.0025
24MiniMax M3$0.0027
25LongCat 2.0$0.0032
26Qwen3.7 Plus$0.0033
27Gemini 3.5 Flash Lite$0.0034
28Qwen3.6 35B A3B$0.0034
29Qwen3.8 27B$0.0049

Input Reasoning Answer

Cost per question, as billed

All models · lower is better

1Granite 4.0 Micro$0.000051
2gpt-oss-20b$0.000066
3Granite 4.2 8B$0.000077
4Nova Micro 1.0$0.000087
5Mercury 2.5$0.00011
6Phi 4$0.00012
7gpt-oss-120b$0.00013
8GLM 5.3 Flash$0.00015
9Ministral 3 3B 2512$0.00019
10GPT-6 Luna$0.00025
32Gemma 4 31B$0.0014

61 more models · see the full leaderboard

Speed and latency

Median of 3 streamed runs of the same prompt, measured 25 Oct 2026.

Output speed

Peer group ($0.15 to $1) · higher is better

16 more models · see the full leaderboard

Time to first token

Peer group ($0.15 to $1) · lower is better

16 more models · see the full leaderboard

End-to-end response time

Peer group ($0.15 to $1) · lower is better

16 more models · see the full leaderboard

Latency vs. output speed

Gemma 4 31B circled · quick to start and fast sits top left

Results by category

Questions answered correctly in each category

Science11/30

Engineering0/15

Medicine2/11

Economics5/9

Law1/6

Humanities & social science5/9

More from Google

Google#2Gemini 4 Argon84.8% verification score69 tok/s3.3 s first token$0.0063 / question

Google#3Gemini 3.8 Flash82.5% verification score142 tok/s3.2 s first token$0.0037 / question

Google#9Gemini 3.1 Pro Preview77.5% verification score124 tok/s3.4 s first token$0.017 / question

All Google models →

Compare Gemma 4 31B

Head to head with its closest rivals, other Google models and the top models overall.

Hosts and prices

Every host serving this model through OpenRouter, from OpenRouter's catalogue

Host	Input / 1M	Output / 1M	Context	Max output	Quantization
DeepInfra	$0.16	$0.32	262k	236k	fp8
CoreWeave	$0.1	$0.34	262k	236k	fp4
Venice	$0.12	$0.36	256k	8k	fp4
Chutes	$0.12	$0.37	131k	66k	fp4
DekaLLM	$0.14	$0.4	262k	236k	—
Crusoe	$0.14	$0.4	262k	262k	bf16
Friendli	$0.14	$0.4	262k	8k	—
Novita	$0.14	$0.4	262k	131k	bf16
Parasail	$0.15	$0.4	262k	236k	fp8
ModelRun	$0.75	$1	262k	236k	fp4
SiliconFlow	$0.75	$1	262k	236k	fp8
Io Net	$0.361	$1.0925	262k	16k	—
SambaNova	$0.38	$1.15	131k	118k	—

Sources

Scores, speed and billed cost are our own measurements. Prices, context window and hosts come from OpenRouter's catalogue on the test date.

Gemma 4 31B on OpenRouter: list prices and hosts
Model card on Hugging Face: weights and licence
SuperGPQA (paper): 50 of the questions, ODC-BY 1.0
MMLU-Pro (paper): 30 of the questions, MIT
OpenRouter generation records: where billed cost and native token counts come from
Our methodology and editorial standards

Gemma 4 31B: frequently asked questions

How accurate is Gemma 4 31B?: Gemma 4 31B answered 24 of 80 questions correctly (30.0%) in our 2026-10-27 run, ranking #64 of 78 models.
How fast is Gemma 4 31B?: Gemma 4 31B generated a median of 58 output tokens per second in our 2026-10-27 run, ranking #54 of 78. Median time to first token was 0.6 s.
What is the latency of Gemma 4 31B?: Gemma 4 31B had a median time to first token of 0.6 s in our 2026-10-27 run, ranking #26 of 78 models.
How much does Gemma 4 31B cost?: Gemma 4 31B cost $0.0014 per question on average as billed in our 2026-10-27 run, reasoning tokens included.
What is Gemma 4 31B API pricing?: Gemma 4 31B is listed at $0.2 per million input tokens and $0.4 per million output tokens on OpenRouter.
What is the context window of Gemma 4 31B?: Gemma 4 31B accepts up to 262,144 tokens of context, according to OpenRouter's model catalogue.
Is Gemma 4 31B open source?: Yes. Gemma 4 31B is an open-weights model: Google publishes the weights, so you can run it yourself as well as through an API.
Who makes Gemma 4 31B?: Gemma 4 31B is made by Google. We tested it through OpenRouter using the model ID google/gemma-4-31b-it.
Where can I use Gemma 4 31B via API?: Through OpenRouter, Gemma 4 31B is served by 3 hosts: CoreWeave, DeepInfra, ModelRun.