Gemma 3 12B vs Gemma 4 31B
Pick up to 3 models · pricing, context and capabilities · simulate your product cost
Data updated on September 30, 2026· 217 models
Gemma 3 introduces multimodality, supporting vision-language input and text outputs.
- Reasoning estimated
- Speed estimated
- ▲
- Input
- Output
- Context
- 131 K tokens
- Max output
- 16 K
- Input, $ per 1M tokens
- ▲$0.050
- Output, $ per 1M tokens
- ▲$0.150
- Cache read, $ per 1M
- —
- Batch (input / output)
- —
- Tool calling
- No
- Released
- 12/01/2024
- Knowledge cutoff
- 2024-12
- API id
- google/gemma-3-12b-it
Gemma 4 31B Instruct is Google DeepMind's 30.7B dense multimodal model supporting text and image input with text output.
- Reasoning estimated
- ▲
- Speed estimated
- Input
- Output
- Context
- ▲262 K tokens
- Max output
- 16 K
- Input, $ per 1M tokens
- $0.090
- Output, $ per 1M tokens
- $0.340
- Cache read, $ per 1M
- $0.050
- Batch (input / output)
- —
- Tool calling
- Yes
- Released
- ▲04/02/2026
- Knowledge cutoff
- —
- API id
- google/gemma-4-31b-it
Which is the better API choice: Gemma 3 12B vs Gemma 4 31B?
- Gemma 3 12B: cheapest input (44% less than Gemma 4 31B)
- Gemma 3 12B: cheapest output
- Gemma 4 31B: largest context (262 K)
- Gemma 4 31B: the newest (04/02/2026)
- Gemma 4 31B: most reasoning (estimated)
- Gemma 3 12B: the fastest (estimated)
Gemma 3 12B costs 44% less on input and 56% less on output than Gemma 4 31B ($0.050 / $0.150 versus $0.090 / $0.340 per million tokens). Gemma 4 31B accepts more context (262 K). On capabilities, Gemma 4 31B has a reasoning mode.
What would your product cost?
Estimate the monthly cost with your app's real volume. List prices, no cache or batch discounts.
With Gemma 3 12B you save $1.36 a month versus Gemma 4 31B (50% less).
The 6 cheapest for this scenario (chat, last 12 months)
- 1Lyria 3 Pro Preview Google$0.00
- 2Lyria 3 Clip Preview Google$0.00
- 3Qwen3.7 Flash Qwen$0.97
- 4DeepSeek V4 Flash 0731 DeepSeek$1.55
- 5Ministral 3 3B 2512 Mistral$1.90
- 6Qwen3.5-Flash Qwen$2.02
Paste your prompt: tokens and cost per model
Rough estimate (about 4 characters per token). Nothing is sent to any server.
Cost of sending it 1,000 times (input only)
- 1Gemma 3 12B Google$0.004
- 2Gemma 4 31B Google$0.007
- 3Claude Sonnet 5.5 Anthropic$0.152
- 4Claude Sonnet 5 Anthropic$0.152
- 5Claude Opus 5.5 Anthropic$0.304
- 6Claude Opus 5 Anthropic$0.380
- 7Claude Fable 5.1 Anthropic$0.760
- 8Claude Fable 5 Anthropic$0.760
Popular comparisons
- Claude Sonnet 5.5 vs GPT-6.1 Sol
- Claude Sonnet 5.5 vs Gemini 3.8 Flash
- Claude Sonnet 5.5 vs DeepSeek V4.1 Flash
- Claude Sonnet 5.5 vs Grok 4.7
- GPT-6.1 Sol vs Gemini 3.8 Flash
- GPT-6.1 Sol vs DeepSeek V4.1 Flash
- GPT-6.1 Sol vs Grok 4.7
- Gemini 3.8 Flash vs DeepSeek V4.1 Flash
- Gemini 3.8 Flash vs Grok 4.7
- DeepSeek V4.1 Flash vs Grok 4.7
Frequently asked questions
Which is the better API choice: Gemma 3 12B vs Gemma 4 31B?
Gemma 3 12B costs 44% less on input and 56% less on output than Gemma 4 31B ($0.050 / $0.150 versus $0.090 / $0.340 per million tokens). Gemma 4 31B accepts more context (262 K). On capabilities, Gemma 4 31B has a reasoning mode.
Which is cheaper: Gemma 3 12B and Gemma 4 31B?
Gemma 3 12B is cheaper: $0.050 per million input tokens and $0.150 per million output tokens, versus $0.090 / $0.340 for Gemma 4 31B.
Which has more context: Gemma 3 12B and Gemma 4 31B?
Gemma 4 31B accepts 262 K tokens of context; Gemma 3 12B accepts 131 K.
Which is newer: Gemma 3 12B and Gemma 4 31B?
Gemma 4 31B was released on 04/02/2026; Gemma 3 12B on 12/01/2024.
What would 10,000 requests a month cost with Gemma 3 12B and Gemma 4 31B?
With 1,500 input and 400 output tokens per request: Gemma 3 12B: $1.35; Gemma 4 31B: $2.71 a month, at list price.
Where do the prices come from and how often are they updated?
From the public catalogs of OpenRouter, models.dev and LiteLLM, which publish each provider's list prices. They sync once a day and the last update date is shown above the comparison.
What does price per million tokens mean?
It is what the provider charges per million tokens you send (input) or the model generates (output). A token is roughly 4 characters. Output almost always costs more than input.
Sources: OpenRouter, models.dev and LiteLLM. Prices in USD per million tokens, list price, no volume discounts. Prompt tokens are an estimate; each provider's tokenizer may differ.