AI21-Jamba-Reasoning-3B

AI21-Jamba-Reasoning-3B (AI21 Labs)

Small reasoning-focused language model built to think step by step.

• Reasoning-first design. A ~3B-parameter model from the Jamba family, trained with an emphasis on logical reasoning, structured thinking, and multi-step problem solving.

• Strong at “thinking” tasks. Performs well on math, logic, symbolic reasoning, and tasks that benefit from explicit intermediate steps rather than fluent text alone.

• Compact but deliberate. Much smaller than large reasoning models, yet more methodical than general-purpose LLMs of similar size.

• Instruction-aware. Follows prompts reliably and produces clearer reasoning traces compared to standard chat-focused models.

• Practical footprint. Suitable for single-GPU or optimized CPU setups where reasoning quality matters more than raw generation speed.

Why pick it for Norman AI?

AI21-Jamba-Reasoning-3B is a good choice when you want reasoning behavior without heavyweight models. It fits evaluations, agent logic, and structured decision workflows where correctness and clarity matter more than verbosity.

messages = [
    {"role": "system", "content": "You are a helpful AI assistant."},
    {"role": "user", "content": "Can you provide ways to eat combinations of bananas and dragonfruits?"},
    {"role": "assistant",
     "content": "Sure! Here are some ways to eat bananas and dragonfruits together"},
    {"role": "user", "content": "What about solving an 2x + 3 = 7 equation?"},
]

response = await norman.invoke(
    {
        "model_name": "ai21-jamba-reasoning-3b",
        "inputs": [
            {
                "display_title": "Prompt",
                "data": messages
            }
        ]
    }
)