Juno.

    The AI model layer.

    Explore modelsHugging Face

    Juno Series

    Juno-N

    Locai One

    Built on Nemotron. Quantised to NVFP4 and specialised with our SPACE compression algorithm for software engineering and agentic workloads. Ships at launch.

    Hugging FaceJuno-N

    Juno-Q

    Locai One

    Built on Qwen. The multilingual option, strong across languages and code.

    Coming soon to Hugging Face

    Juno-G

    Locai One

    Built on Gemma. Built on Gemma for instruction following, safety and code.

    Coming soon to Hugging Face

    Juno-I

    Locai One Pro

    Built on Inkling. A second state-of-the-art base for teams wanting an alternative to Juno-D.

    Coming soon to Hugging Face

    Juno-D

    Locai One Pro

    Built on DeepSeek V4 Flash. For deep analysis and long documents.

    Coming soon to Hugging Face

    Benchmarks

    100.3%

    Of base · Targeted benchmarks

    85.2%

    Of base · Outside target

    −19%

    Parameters vs base

    BenchmarkNemotron 3.5 LightningJuno-N

    Targeted · software engineering & agentic

    LiveCodeBench v6
    77.97
    77.09
    HumanEval+
    89.33
    90.55
    MBPP+
    81.49
    80.29
    SciCode (subtask)
    36.09
    36.98

    Outside target · general knowledge

    MMLU Redux
    90.00
    82.18
    SimpleQA (rubric)
    47.95
    37.88

    Juno-N retains performance where we deliberately targeted it, while its capability profile is reduced outside those target areas. That narrower profile is intentional.