
Agent Platform Scorecard
Twelve dimensions. Seven providers. Re-graded as announcements land — with the market's reaction alongside.
The assistant most people mean when they say 'AI'.
Still the default consumer assistant and a serious agent contender via agent mode and a deep API platform. Owns its models but leans on Microsoft for cloud and enterprise distribution. Weakest on the device edge.
Frontier reasoning and the assistant most users measure everyone else against.
Agent mode and Operator-style browsing ship real multi-step work, not just demos.
Advanced Voice set the bar for latency, interruption handling, and natural turn-taking.
Connectors and actions are expanding, but it reaches into apps from the outside — not the OS.
Persistent memory across chats is mature; lacks deep device/calendar/file grounding.
GPTs, connectors, and a growing app surface, but no open cross-vendor protocol of its own.
The reference developer platform: tool use, structured output, Assistants, fine-tuning, scale.
Owns the frontier model stack end to end; compute via Microsoft and increasingly its own.
Massive distribution plus an explicit agent push position it as a leading agent platform.
Essentially no device or OS-level local-inference story of its own.
ChatGPT Enterprise and the Azure relationship give a real, fast-growing enterprise motion.
Highly capable, but a hosted assistant — not yet a governable, revocable control plane with durable memory.
From the default consumer assistant toward a full agent platform — the open flank is the device edge and a hardware surface of its own.
Frontier models, the best-known API, and a maturing agent mode — but no device story.
Deeper agent infrastructure and a rumored dedicated device would attack the on-device gap that caps its scorecard.