Skip to main content

​
Available models

Model IDParametersContextSpeedBest For
deepseek-v3671B (MoE)160KMediumGeneral, coding, best value
deepseek-r1671B (MoE)64KSlowReasoning, math, analysis

​
DeepSeek V3 vs R1 — When to use which?

DeepSeek V3DeepSeek R1
TypeGeneral-purposeReasoning (chain-of-thought)
SpeedMediumSlower (thinks step-by-step)
Context160K tokens64K tokens
Best forCoding, writing, general tasksMath, logic, complex analysis
PriceLowerHigher (more output tokens from reasoning)
Comparable toGPT-5 classo1 class (96% cheaper)
Rule of thumb: Use V3 by default. Switch to R1 only when you need deep reasoning — math proofs, logic puzzles, multi-step analysis, or research.

​
Recommendation

Use deepseek-v3 as your default — it’s GPT-5 class quality at the best price on Kyma. The most tokens processed per user on the platform. Use deepseek-r1 for reasoning tasks — when you need step-by-step thinking. It maps to the reasoning model alias. 
from openai import OpenAI

client = OpenAI(base_url="https://kymaapi.com/v1", api_key="ky-...")

# General tasks — use V3
response = client.chat.completions.create(
    model="deepseek-v3",
    messages=[{"role": "user", "content": "Explain quantum computing simply"}]
)

# Reasoning tasks — use R1
response = client.chat.completions.create(
    model="deepseek-r1",
    messages=[{"role": "user", "content": "Prove that sqrt(2) is irrational"}]
)

​
Model aliases

AliasResolves to
reasoningdeepseek-r1
# Using alias
model="reasoning"  # → deepseek-r1