Insight

    See which AI products are available, who makes them, and how the alternatives compare.

    What you can use. Who makes it.

    Start with an AI family or a specialist tool, then compare alternatives and the company behind each one. These are selected entry points into the catalog, not a ranking of every available product.

    Each family brings its assistant, models and coding tool together. Open one page to choose how to use it.

    ChatGPT app, GPT models and Codex in one AI family

    Strength

    92.2

    3/3 rated

    Claude app, Claude models and Claude Code in one AI family

    Strength

    91.2

    3/3 rated

    Google’s AI apps, models, coding tools and enterprise platform in one family

    Strength

    91.2

    3/3 rated

    Korean companies

    177 profiles · products & portfolios

    Global companies

    379 profiles · products & portfolios

    Chinese companies

    41 profiles · products & portfolios

    Strength is the unweighted geometric mean of product value, adoption and durability, displayed on 0–100. ≈ marks editorial ratings; open a product to inspect the rationale, comparison and sources.

    Map the ecosystem

    The Agentic AI Stack

    Break down today's ecosystem: what each layer does, who operates there, and where the strengths and gaps are. Compare global, Chinese and Korean players across the current agentic-AI stack.

    Start with the stack map
    1. L5
      ApplicationsDigital work and physical robots
    2. L4
      AgentsPlan, use tools, execute
    3. L3
      InfrastructureServe and operate AI
    4. L2
      ModelsThe capability engine
    5. L1
      HardwareCompute and memory

    Follow the bigger picture

    The Five Waves

    Now place that stack in the bigger story. Compute, models, agents, science and physical AI are overlapping waves of change. We are in Wave 3, with room to grow. Science and physical AI will need new stack maps of their own.

    1. 01

      Compute & memory

      The physical foundation

    2. 02

      Frontier models

      The capability race

    3. 03

      Agentic AI

      Now · still expanding

    4. 04

      AI for science

      Hypotheses meet experiments

    5. 05

      Physical AI

      A longer-term direction

    Then explore the five waves

    Evidence & updates

    2026-10-22 · Reviewed developments and editorial interpretation

    What changes the decision

    Customer ownership matters more as models improve

    Our inference: the useful moat question is whether customers keep a workflow when its underlying model changes. Harvey has disclosed organizational deployment; product retention and inference margins still need separate evidence.

    Open weights move the cost question to deployment

    Beam, Muse Spark Open and Step 5 Preview now have weight releases. Sparse models still need memory for all weights. Compare quality-adjusted throughput, serving reliability and licence terms before choosing a self-hosted route.

    Specialization needs a measurable reason to exist

    Timestral targets numerical forecasting on CPUs, while Tinker makes post-training accessible. These are candidates to test against current baselines. A narrow benchmark lead or a specialist label alone does not establish durable value.

    Check retirement dates as carefully as launches

    Solar Pro 2 and Pro 3 are scheduled to end service on October 30 KST. Solar Pro 4 and Mini 4 are the current alternatives. Test tool calls and retries before migration; a pricing table does not tell you the cost of a completed task.

    Tools to test the picture