AI Security

    Kimi K2.5: benchmark labels now distinguish AIME 2025 and Terminal-Bench 2.0

    Tracker confirms version-specific labels for two benchmarks

    Benchmark labels now distinguish AIME 2025 and Terminal-Bench 2.0 — verified by the OnAirToday AI Progress Tracker against Moonshot's published documentation.

    O

    OnAirToday

    Newsroom

    OnAirToday

    The OnAirToday AI Progress Tracker recorded a verified change to Moonshot's Kimi K2.5 (version kimi-k2.5): Benchmark labels now distinguish AIME 2025 and Terminal-Bench 2.0.

    Kimi K2.5 is Moonshot's multimodal model for reasoning, coding, visual understanding and tool-assisted workflows, with a documented 256K-token context window. Its published benchmark results distinguish tool-assisted performance from unaided reasoning and do not establish reliability for unsupervised consequential operations.

    What changed in this review:

    • Added attributed security reports, distinguishing Kimi K2.5 coverage from reports concerning Kimi K3 and Kimi K2.6.
    • A family-level Kimi K3 Redis exploit report does not establish a Kimi K2.5 vulnerability.
    • Current model-specific Arena and SimpleQA results: Not yet confirmed by public reporting.
    • Capability score 67 → 70
    • Flaws and limitations updated
    • Risks and threats updated

    Anyone building on Kimi K2.5 should re-check the model's risk dashboard before shipping, since flaws and reported incidents are what moved in this review.

    The full profile — capabilities, benchmark citations, known flaws and risk guidance — is on the tracker at /ai-models/kimi-k2-5.

    References:

    • https://huggingface.co/moonshotai/Kimi-K2.5
    • https://platform.moonshot.ai/docs/pricing/chat
    • https://huggingface.co/moonshotai/Kimi-K2-Thinking
    AI Analysis

    Generated insight by OnAirToday AI

    Version-specific benchmark labels make it clearer which evaluations are being referenced. This can help readers interpret benchmark results without conflating different test versions.

    Powered by AI • Not a substitute for human analysisMore Insights →
    View the primary source →
    Share
    Filed under:AI Security

    Comments

    Keep comments relevant and respectful. No links, advertising, self-promotion, or spam.

    0/1000