DeepSeek V3.1on Krater.ai
DeepSeek-V3.1 is a large hybrid reasoning model (671B parameters, 37B active) that supports both thinking and non-thinking modes via prompt templates. It extends the DeepSeek-V3 base with a two-phase long-context
Released August 21, 2025
Specifications
DeepSeek V3.1 at a glance
Context
163.8K tokens
Max output
32.8K tokens
Input
Text
Output
Text
Input price
$0.25 / 1M
Output price
$0.95 / 1M
Knowledge cutoff
2025-03-31
Capabilities
What it is good at
Catalog capabilities
- Generates text responses
- Supports tool calling
- Supports structured outputs
- Instruction format: deepseek-v3.1
Good for
Benchmarks
Catalog data, not Krater benchmarks.
Design Arena 3d rank
#71
Design Arena gamedev rank
#71
Compare
Compare DeepSeek V3.1
DeepSeek V3.1 vs DeepSeek V3.2
DeepSeek-V3.2 is a large language model designed to harmonize high computational efficiency with strong reasoning and agentic tool-use performance.
Open comparisonDeepSeek V3.1 vs DeepSeek V3 0324
V3, a 685B-parameter, mixture-of-experts model, is the latest iteration of the flagship chat model family from the DeepSeek team.
Open comparisonDeepSeek V3.1 vs DeepSeek V4 Flash 0731
V4 Flash 0731 is a sparse mixture-of-experts model from DeepSeek, with 13B active parameters out of 284B total.
Open comparisonFAQ
Questions about DeepSeek V3.1
Clear answers based on the model catalog and Krater access.
Use DeepSeek V3.1 in your API
Connect with Krater's unified API.