The AI arena is free today

Open Superagent

DeepSeek-V2.5 vs Step3-VL-10B

Step3-VL-10B leads the LLM Stats Score 25.5 to 8.1.

DeepSeek · StepFun · Updated for 2026

Which is better?

Step3-VL-10B leads the overall LLM Stats Score 25.5 to 8.1, ranking #155 overall.

Based on current LLM Stats indexes, shared benchmarks, pricing, and model metadata for 2026.

Choose DeepSeek-V2.5

  • you want predictable pricing at $0.14/M input and $0.28/M output

Choose Step3-VL-10B

  • overall performance matters — it scores 25.5 and ranks #155 on LLM Stats
  • your work emphasizes reasoning — it leads those capability indexes
  • you want the most recent training data — it shipped Jan 2026

At a glance

The differences that matter most.

DeepSeek-V2.5
DeepSeek
Step3-VL-10B
StepFun
Core performance indexes
LLM Stats Score
8.1
#280
25.5
#155
Reasoning
8.2
#276
25.4
#150
Cost, coverage & limits
Benchmark wins
—
—
Input price
$0.14 / M
— / M
Output price
$0.28 / M
— / M
Context window
8,192
—

Capability indexes

Additional strengths measured across groups of related public benchmarks

1 shared
Index
DeepSeek-V2.5
Step3-VL-10B
Math
14.0#222
23.5#126
Conservative TrueSkill rating · higher is betterHow scores work

Individual benchmarks

15 reported for DeepSeek-V2.5 · 6 for Step3-VL-10B

No common benchmarks found

DeepSeek-V2.5 and Step3-VL-10Bdon't have any common benchmark datasets to compare. They may have been evaluated on different testing suites.

Human preference

Blind head-to-head votes and playground preference scores

Model Size

Parameter count comparison

226.0B diff

DeepSeek-V2.5 has 226.0B more parameters than Step3-VL-10B, making it 2260.0% larger.

DeepSeek
DeepSeek-V2.5
236.0Bparameters
StepFun
Step3-VL-10B
10.0Bparameters
236.0B
DeepSeek-V2.5
10.0B
Step3-VL-10B

Context Window

Maximum input and output token capacity

Only DeepSeek-V2.5 specifies input context (8,192 tokens). Only DeepSeek-V2.5 specifies output context (8,192 tokens).

DeepSeek
DeepSeek-V2.5
Input8,192 tokens
Output8,192 tokens
StepFun
Step3-VL-10B
Input- tokens
Output- tokens
Sun Sep 13 2026 • llm-stats.com

Input capabilities

Documented input modalities across available providers

Step3-VL-10B supports multimodal inputs, whereas DeepSeek-V2.5 does not.

Step3-VL-10B can handle both text and other forms of data like images, making it suitable for multimodal applications.

DeepSeek-V2.5

Text
Images
Audio
Video

Step3-VL-10B

Text
Images
Audio
Video

License

Usage and distribution terms

DeepSeek-V2.5 is licensed under deepseek, while Step3-VL-10B uses Apache 2.0.

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

DeepSeek-V2.5

deepseek

Open weights

Step3-VL-10B

Apache 2.0

Open weights

Release Timeline

When each model was launched

DeepSeek-V2.5 was released on 2024-05-08, while Step3-VL-10B was released on 2026-01-15.

Step3-VL-10B is 21 months newer than DeepSeek-V2.5.

DeepSeek-V2.5

May 8, 2024

2.4 years ago

Step3-VL-10B

Jan 15, 2026

8 months ago

1.7yr newer

Knowledge Cutoff

When training data ends

Neither model specifies a knowledge cutoff date.

Unable to compare the recency of their training data.

No cutoff dates available

Outputs Comparison

Notice missing or incorrect data?Start an Issue discussion→

Judge for yourself.

Run your own prompts against DeepSeek-V2.5 and Step3-VL-10B side-by-side, then vote on the output you prefer.

DeepSeek-V2.5
✓ Preferred
Step3-VL-10B
Open in Playground

FAQ

Common questions about DeepSeek-V2.5 vs Step3-VL-10B.

Which is better, DeepSeek-V2.5 or Step3-VL-10B?

Step3-VL-10B leads the LLM Stats Score 25.5 to 8.1. DeepSeek-V2.5 is made by DeepSeek and Step3-VL-10B is made by StepFun. The best choice depends on your use case — compare their capability indexes, individual benchmarks, pricing, and limits above.

How does DeepSeek-V2.5 compare to Step3-VL-10B in benchmarks?

DeepSeek-V2.5 scores GSM8k: 95.1%, MT-Bench: 90.2%, HumanEval: 89.0%, BBH: 84.3%, AlignBench: 80.4%. Step3-VL-10B scores MMBench: 91.8%, AIME 2025: 87.7%, MathVista: 84.0%, MMMU: 78.1%, MathVision: 70.8%.

What are the context window sizes for DeepSeek-V2.5 and Step3-VL-10B?

DeepSeek-V2.5 supports 8K tokens and Step3-VL-10B supports an unknown number of tokens. A larger context window lets you process longer documents, conversations, or codebases in a single request.

What are the main differences between DeepSeek-V2.5 and Step3-VL-10B?

Key differences include LLM Stats Score (8.1 vs 25.5), multimodal support (no vs yes), licensing (deepseek vs Apache 2.0). See the full comparison above for benchmark-by-benchmark results.

Who makes DeepSeek-V2.5 and Step3-VL-10B?

DeepSeek-V2.5 is developed by DeepSeek and Step3-VL-10B is developed by StepFun.

Related comparisons