JetBrains Mellum2 12B A2.5B inference overview

Price per 1M tokens

$.05 (input)
$.10 (output)

Parameters

2.5B (active)
12B (total)

Context Window

131K

Release Date

Jun 2026

JetBrains Mellum2 12B A2.5B inference details

Mellum2-12B-A2.5B-Instruct is a JetBrains-built MoE model with 12B total parameters and only 2.5B active per token, supporting a 131K context window. Trained from scratch on natural language and code, it is designed as a fast, cost-efficient model for production AI systems — excelling at routing, Q&A, tool use, sub-agent tasks, and RAG pipelines where latency and throughput matter most.

Created by: 

JetBrains

License: 

apache-2.0
				
					import openai
import weave

# Weave autopatches OpenAI to log LLM calls to W&B
weave.init("<team>/<project>")

client = openai.OpenAI(
    # The custom base URL points to W&B Inference
    base_url='https://api.inference.wandb.ai/v1',

    # Get your API key from https://wandb.ai/authorize
    # Consider setting it in the environment as OPENAI_API_KEY instead for safety
    api_key="<your-apikey>",

    # Optional: Team and project for usage tracking
    project="<team>/<project>",
)

response = client.chat.completions.create(
    model="JetBrains/Mellum2-12B-A2.5B-Instruct",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Tell me a joke."}
    ],
)

print(response.choices[0].message.content)
				
			

JetBrains Mellum2 12B A2.5B resources

Screenshot 2025-07-30 at 1.03.14 PM
Course
AI engineering course: Agents
Inference_logo
Guide
W&B Inference powered by CoreWeave
Screenshot 2025-07-30 at 8.00.14 AM
Whitepaper
A primer on building successful AI agents