Capability feed

Every model release and benchmark change that re-scores the AI-deliverability frontier, newest first.

  1. Model release
    GPT-5

    400k-token context and native image input cross the threshold for end-to-end return prep.

    Tax PreparersServices-as-SoftwareAccounting
    POV: Services-as-Software
  2. Model release
    GPT-5

    Agentic coding scores jump enough to delegate whole refactors under oversight.

    Software DevelopersAutonomous AgentsSoftware
    POV: Autonomous Agents as digital employees
  3. Model release
    Claude Opus 4.8

    Long-context legal review becomes reliable enough to draft first-pass memos.

    ParalegalsServices-as-SoftwareLegal
    POV: Services-as-Software
  4. Benchmark update
    GDPval

    +12 points on the tax-preparation task family re-scores the AI-deliverability frontier.

    Tax PreparersServices-as-SoftwareGDPval
    POV: Services-as-Software