Model Comparison

DeepSeek-V3.2 (Thinking) vs Gemma 4 31BWhich is better in 2026?

Gemma 4 31B significantly outperforms across most benchmarks. Gemma 4 31B is 1.6x cheaper per token.

Verdict: DeepSeek-V3.2 (Thinking) vs Gemma 4 31B — which is better?

DeepSeek-V3.2 (Thinking) (by DeepSeek) and Gemma 4 31B (by Google) are two of the AI models people compare most. Here is how they stack up on benchmarks, price and capabilities, and which one to pick in 2026.

DeepSeek-V3.2 (Thinking) outperforms in 0 benchmarks, while Gemma 4 31B is better at 4 benchmarks (GPQA, Humanity's Last Exam, MMLU-Pro, t2-bench). Gemma 4 31B significantly outperforms across most benchmarks.

On price, Gemma 4 31B is roughly 1.6x cheaper per token on a blended 3:1 input/output basis, which adds up quickly at production volume.

Gemma 4 31B also accepts a larger context window (262,144 input tokens), making it the stronger choice for long documents and large codebases.

Choose DeepSeek-V3.2 (Thinking) if…

  • you want predictable pricing at $0.28/M input and $0.42/M output

Choose Gemma 4 31B if…

  • you want the strongest raw capability — it leads on 4 of 4 shared benchmarks
  • cost matters — it's about 1.6x cheaper per token
  • you process long inputs — it offers a 262,144 token context window
  • you want the most recent training data — it shipped Apr 2026

Performance Benchmarks

Comparative analysis across standard metrics

4 benchmarks

DeepSeek-V3.2 (Thinking) outperforms in 0 benchmarks, while Gemma 4 31B is better at 4 benchmarks (GPQA, Humanity's Last Exam, MMLU-Pro, t2-bench).

Gemma 4 31B significantly outperforms across most benchmarks.

Tue Jul 21 2026 • llm-stats.com

Arena Performance

Human preference votes

Pricing Analysis

Price comparison per million tokens

Gemma 4 31B costs less

For input processing, DeepSeek-V3.2 (Thinking) ($0.28/1M tokens) is 2.2x more expensive than Gemma 4 31B ($0.13/1M tokens).

For output processing, DeepSeek-V3.2 (Thinking) ($0.42/1M tokens) is 1.1x more expensive than Gemma 4 31B ($0.38/1M tokens).

In conclusion, DeepSeek-V3.2 (Thinking) is more expensive than Gemma 4 31B.*

* Using a 3:1 ratio of input to output tokens

Lowest available price from all providers
Tue Jul 21 2026 • llm-stats.com
DeepSeek
DeepSeek-V3.2 (Thinking)
Input tokens$0.28
Output tokens$0.42
Best providerDeepSeek
Google
Gemma 4 31B
Input tokens$0.13
Output tokens$0.38
Best providerDeepinfra
Notice missing or incorrect data?Start an Issue→

Model Size

Parameter count comparison

654.3B diff

DeepSeek-V3.2 (Thinking) has 654.3B more parameters than Gemma 4 31B, making it 2131.3% larger.

DeepSeek
DeepSeek-V3.2 (Thinking)
685.0Bparameters
Google
Gemma 4 31B
30.7Bparameters
685.0B
DeepSeek-V3.2 (Thinking)
30.7B
Gemma 4 31B

Context Window

Maximum input and output token capacity

Gemma 4 31B accepts 262,144 input tokens compared to DeepSeek-V3.2 (Thinking)'s 131,072 tokens. Gemma 4 31B can generate longer responses up to 131,072 tokens, while DeepSeek-V3.2 (Thinking) is limited to 65,536 tokens.

DeepSeek
DeepSeek-V3.2 (Thinking)
Input131,072 tokens
Output65,536 tokens
Google
Gemma 4 31B
Input262,144 tokens
Output131,072 tokens
Tue Jul 21 2026 • llm-stats.com

Input Capabilities

Supported data types and modalities

Gemma 4 31B supports multimodal inputs, whereas DeepSeek-V3.2 (Thinking) does not.

Gemma 4 31B can handle both text and other forms of data like images, making it suitable for multimodal applications.

DeepSeek-V3.2 (Thinking)

Text
Images
Audio
Video

Gemma 4 31B

Text
Images
Audio
Video

License

Usage and distribution terms

DeepSeek-V3.2 (Thinking) is licensed under MIT, while Gemma 4 31B uses Apache 2.0.

License differences may affect how you can use these models in commercial or open-source projects.

DeepSeek-V3.2 (Thinking)

MIT

Open weights

Gemma 4 31B

Apache 2.0

Open weights

Release Timeline

When each model was launched

DeepSeek-V3.2 (Thinking) was released on 2025-12-01, while Gemma 4 31B was released on 2026-04-02.

Gemma 4 31B is 4 months newer than DeepSeek-V3.2 (Thinking).

DeepSeek-V3.2 (Thinking)

Dec 1, 2025

7 months ago

Gemma 4 31B

Apr 2, 2026

3 months ago

4mo newer

Knowledge Cutoff

When training data ends

Gemma 4 31B has a documented knowledge cutoff of 2025-01-01, while DeepSeek-V3.2 (Thinking)'s cutoff date is not specified.

We can confirm Gemma 4 31B's training data extends to 2025-01-01, but cannot make a direct comparison without DeepSeek-V3.2 (Thinking)'s cutoff date.

DeepSeek-V3.2 (Thinking)

—

Gemma 4 31B

Jan 2025

Provider Availability

DeepSeek-V3.2 (Thinking) is available from DeepSeek. Gemma 4 31B is available from DeepInfra, FriendliAI, Novita, Together.

DeepSeek-V3.2 (Thinking)

deepseek logo
DeepSeek
Input Price:Input: $0.28/1MOutput Price:Output: $0.42/1M

Gemma 4 31B

deepinfra logo
Deepinfra
Input Price:Input: $0.13/1MOutput Price:Output: $0.38/1M
friendli logo
FriendliAI
Input Price:Input: $0.14/1MOutput Price:Output: $0.40/1M
novita logo
Novita
Input Price:Input: $0.14/1MOutput Price:Output: $0.40/1M
together logo
Together
Input Price:Input: $0.39/1MOutput Price:Output: $0.97/1M
* Prices shown are per million tokens

Outputs Comparison

Notice missing or incorrect data?Start an Issue discussion→

Key Takeaways

DeepSeek

No standout differentiators in the data we have for this pair.

Google
Larger context window (262,144 tokens)
Supports multimodal inputs
Less expensive input tokens
Less expensive output tokens
Higher GPQA score (84.3% vs 82.4%)
Higher Humanity's Last Exam score (26.5% vs 25.1%)
Higher MMLU-Pro score (85.2% vs 85.0%)
Higher t2-bench score (86.4% vs 80.2%)

Detailed Comparison

Interactive Arena

Judge for yourself.

Run your own prompts against DeepSeek-V3.2 (Thinking) and Gemma 4 31B side-by-side, then vote on the output you prefer.

DeepSeek-V3.2 (Thinking)
✓ Preferred
Gemma 4 31B
Open in Playground
AI Model Comparison Table
Feature
DeepSeek
DeepSeek-V3.2 (Thinking)
Google
Gemma 4 31B

FAQ

Common questions about DeepSeek-V3.2 (Thinking) vs Gemma 4 31B.

Which is better, DeepSeek-V3.2 (Thinking) or Gemma 4 31B?

Gemma 4 31B significantly outperforms across most benchmarks. DeepSeek-V3.2 (Thinking) is made by DeepSeek and Gemma 4 31B is made by Google. The best choice depends on your use case — compare their benchmark scores, pricing, and capabilities above.

How does DeepSeek-V3.2 (Thinking) compare to Gemma 4 31B in benchmarks?

DeepSeek-V3.2 (Thinking) scores AIME 2025: 93.1%, HMMT 2025: 90.2%, MMLU-Pro: 85.0%, LiveCodeBench: 83.3%, GPQA: 82.4%. Gemma 4 31B scores AIME 2026: 89.2%, MMMLU: 88.4%, t2-bench: 86.4%, MathVision: 85.6%, MMLU-Pro: 85.2%.

Is DeepSeek-V3.2 (Thinking) cheaper than Gemma 4 31B?

Gemma 4 31B is 2.2x cheaper for input tokens. DeepSeek-V3.2 (Thinking) costs $0.28/M input and $0.42/M output via deepseek. Gemma 4 31B costs $0.13/M input and $0.38/M output via deepinfra.

What are the context window sizes for DeepSeek-V3.2 (Thinking) and Gemma 4 31B?

DeepSeek-V3.2 (Thinking) supports 131K tokens and Gemma 4 31B supports 262K tokens. A larger context window lets you process longer documents, conversations, or codebases in a single request.

What are the main differences between DeepSeek-V3.2 (Thinking) and Gemma 4 31B?

Key differences include context window (131K vs 262K), input pricing ($0.28 vs $0.13/M), multimodal support (no vs yes), licensing (MIT vs Apache 2.0). See the full comparison above for benchmark-by-benchmark results.

Who makes DeepSeek-V3.2 (Thinking) and Gemma 4 31B?

DeepSeek-V3.2 (Thinking) is developed by DeepSeek and Gemma 4 31B is developed by Google.

Related comparisons