plugsky-micro — the reasoning workhorse

Summary

Model	Identity	Context	Latency*	Best for
plugsky-micro	Nemotron 3 Nano (MoE)	1M tokens	~0.55s median	Reasoning tasks, RAG with huge corpora, agent memory

*Median measured latency on the live API at https://api.plugsky.com/v1.

Specs

Property	Value
Model family	Nemotron 3 Nano (MoE)
Parameter class	~9B active / 30B MoE
Context window	1M tokens
Measured latency	~0.55s median on the live API
Capabilities	Reasoning, tool/function calling, long-context
Available on	Hobby plan and above

When to use plugsky-micro

• Reasoning-heavy prompts — math, logic, structured output.
• Long documents — 1M context handles entire codebases and corpora.
• Agents — tool calling with memory that does not forget.
• RAG — retrieve everything, ask once.

Use it

from openai import OpenAI
client = OpenAI(base_url="https://api.plugsky.com/v1", api_key="sk-live-...")
r = client.chat.completions.create(
    model="plugsky-micro",
    messages=[{"role": "user", "content": "Explain 1M context in one sentence."}],
)
print(r.choices[0].message.content)

FAQ

Is plugsky-micro good for production?

Yes — it is one of the most-used models on the platform, with automatic fallback routing when upstream is overloaded.

How does it compare to plugsky-lite?

plugsky-lite is faster (~0.24s) and cheaper for simple tasks; plugsky-micro brings stronger reasoning and 1M context. Both support tools.

Does it support Arabic?

Yes — strong multilingual performance across the catalog. See Arabic LLM.