Google · Open weights
Gemma 4 31B Thinking
Google's Gemma 4 31B instruction-tuned model with thinking explicitly enabled, exposing reasoning traces for complex multimodal and coding workflows. Requests containing video cost $0.14 per million input tokens and $0.40 per million output tokens.
- Reasoning
- Tools
- Vision
- Structured output
- Video input
- Input
- Text, Images, Video
- Output
- Text
- Discovery tier
- Open weights
Pricing
USD per 1M tokens- Input
- $0.10
- Output
- $0.34
- Cache read
- $0.10
Reference prices as published by the model provider. API example
Specifications
- Context window
- 262.1K
- Max output
- 131.1K
- Average output (7 days)
- 1.1K tokens
- Longer than 74% of 468 models
Side by side
Gemma 4 31B Thinking compared
Specifications, prices, speed and benchmark scores next to each other. Benchmarks below show every compared model too.
| Gemma 4 31B ThinkingGoogleThis model | Gemma 3 12B ITGoogleRemove | |
|---|---|---|
| Provider | ||
| Discovery tier | Open weights | Open weights |
| Added | Apr 2, 2026 | Jan 1, 2024 |
| Limits | ||
| Context window | 262.14KBest | 131.07K |
| Max output | 131.07KBest | 16.38K |
| Average output | 1.1K tokens | 168 tokens |
| Pricing, USD per 1M tokens | ||
| Input | $0.10Best | $0.27 |
| Output | $0.34 | $0.27Best |
| Cache read | $0.10 | — |
| Performance | ||
| Throughput | 27.7 tok/s | — |
| Time to first token | 7.2 s | — |
| Request success (24h) | 99.7% | 100.0%Best |
| Benchmark indexes | ||
| Intelligence Index | 14.7Best | 3.8 |
| Coding Index | 43.4Best | 5.8 |
| Math Index | — | 18.3 |
| Capabilities | ||
| Reasoning | Yes | No |
| Tools | Yes | No |
| Vision | Yes | No |
| Structured output | Yes | No |
| PDF input | No | Yes |
| Video input | Yes | No |
| Reasoning effort | — | — |
Public evaluations
Benchmarks
Independent evaluation results, with how each score ranks among the models in the Noumen catalog.
- Gemma 4 31B Thinking
- Gemma 3 12B IT
Intelligence Index
- Gemma 4 31B Thinking14.7
- Gemma 3 12B IT3.8
Gemma 4 31B Thinking: Better than 35% of 326 scored models
Indexes
Coding Index
Composite of the coding evaluations
- Gemma 4 31B Thinking: 43.4Top
- Gemma 3 12B IT: 5.8
Math Index
Composite of the math evaluations
- Gemma 4 31B Thinking: —
- Gemma 3 12B IT: 18.3
Agentic work
AutomationBench-AA
Workflow automation, penalized for guardrail violations
- Gemma 4 31B Thinking: 4.9%Top
- Gemma 3 12B IT: 0.2%
AutomationBench-AA tasks completed
Workflows finished end to end without a guardrail violation
- Gemma 4 31B Thinking: 0.5%Top
- Gemma 3 12B IT: 0.0%
AA-Briefcasev1.1
Agentic knowledge work, rated head to head
- Gemma 4 31B Thinking: 368 Elo
- Gemma 3 12B IT: —
GDPval-AAv2.1
Economically valuable occupational tasks, rated head to head
- Gemma 4 31B Thinking: 619 Elo
- Gemma 3 12B IT: —
Document reasoning
GDP.pdf
Professional reasoning over PDF documents, all-pass rate
- Gemma 4 31B Thinking: 6.0%Top
- Gemma 3 12B IT: 1.0%
AA-LCRv1.1
Reasoning across long documents
- Gemma 4 31B Thinking: 69.7%Top
- Gemma 3 12B IT: 8.3%
Reasoning and knowledge
Humanity’s Last Exam
Expert-level questions across academic fields
- Gemma 4 31B Thinking: 23.6%Top
- Gemma 3 12B IT: 4.2%
GPQA Diamond
Graduate-level science questions
- Gemma 4 31B Thinking: 85.7%Top
- Gemma 3 12B IT: 34.9%
MMLU-Pro
Multi-subject knowledge and reasoning
- Gemma 4 31B Thinking: —
- Gemma 3 12B IT: 59.5%
IFBench
Following precise output instructions
- Gemma 4 31B Thinking: 75.6%Top
- Gemma 3 12B IT: 36.7%
Coding
Terminal-Benchv4.0
Practical software engineering in a terminal
- Gemma 4 31B Thinking: 0.0%Top
- Gemma 3 12B IT: 0.0%Top
SciCode
Scientific computing in Python
- Gemma 4 31B Thinking: 45.5%Top
- Gemma 3 12B IT: 16.4%
LiveCodeBench
Recent competitive programming problems
- Gemma 4 31B Thinking: —
- Gemma 3 12B IT: 13.7%
Math
AIME 2025
Competition mathematics, 2025 problems
- Gemma 4 31B Thinking: —
- Gemma 3 12B IT: 18.3%
AIME
Competition mathematics
- Gemma 4 31B Thinking: —
- Gemma 3 12B IT: 22.0%
MATH-500
Competition math problems
- Gemma 4 31B Thinking: —
- Gemma 3 12B IT: 85.3%
Legacy evaluations
Terminal-Bench Hard
Earlier terminal task suite
- Gemma 4 31B Thinking: 36.4%Top
- Gemma 3 12B IT: 0.8%
τ²-Bench Telecom
Tool use in a customer-support simulation
- Gemma 4 31B Thinking: 59.9%Top
- Gemma 3 12B IT: 10.8%
AA-LCR (unversioned)
Long-context reasoning, earlier grading
- Gemma 4 31B Thinking: 69.7%Top
- Gemma 3 12B IT: 8.3%
Scores from Artificial Analysis, updated Oct 26, 2026. Artificial Analysis
Live routing data
Performance and reliability
Recent averages across requests routed for this model. Your own speed depends on prompt length, reasoning effort and load.
- Throughput
- 27.7 tok/s
- Faster than 20% of 254 models
- Time to first token
- 7.2 s
- Quicker than 21% of 254 models
- Prompt cache hits
- 32.2%
- About $0.10 per 1M input tokens with caching
- Request success (24h)
- 99.7%
- Reliable
Cache hits are the share of input tokens recently served from a prompt cache across all requests. They are not a forecast of your own hit rate.
- Healthy (98% or more)
- Degraded (90–98%)
- Down (under 90%)
- No requests
Developers
Use it through the API
Noumen’s API is OpenAI-compatible: any Chat Completions client works with a base URL and a key.
curl https://www.noumen.xyz/api/v1/chat/completions \
-H "Authorization: Bearer $NOUMEN_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "google/gemma-4-31b-it:thinking",
"messages": [{ "role": "user", "content": "Explain this stack trace." }]
}'import os
from openai import OpenAI
client = OpenAI(base_url="https://www.noumen.xyz/api/v1", api_key=os.environ["NOUMEN_API_KEY"])
reply = client.chat.completions.create(
model="google/gemma-4-31b-it:thinking",
messages=[{"role": "user", "content": "Explain this stack trace."}],
)
print(reply.choices[0].message.content)Related models
Other Google models, starting with the same family.