🧠 Thinking Model

Mellum2.1

12B parameters · 2.5B active (MoE) · 131K context — by JetBrains

82.0% LiveCodeBench v6
83.3% AIME 25/26
47.0% SWE-bench Verified
131K Context

⚙️ API Endpoint

Start a vLLM server: vllm serve JetBrains/Mellum2.1-12B-A2.5B-Thinking --reasoning-parser qwen3

Sampling Parameters
Not connected

💬 Chat

Welcome! I'm Mellum2.1 Thinking. I specialize in coding, math, and complex reasoning.

To get started, configure your API endpoint on the left, or try one of the example prompts below.

📊 Benchmarks

Coding

LiveCodeBench v6
82.0
HumanEval+
91.5
MBPP+
79.4

Math

AIME 25/26
83.3
GSM-Plus
88.3

Agentic

SWE-bench Verified
47.0
Terminal-Bench 2.1
17.4

Tool Use

BFCL v4
62.3
WorkBench
44.6