The AI Sift is part of you-do-nothing

← Back to The Latest

OpenAI DevDay ships GPT-6.1 Sol, always-on Dots, and a Decisions API. The agent platform reset happened here

OpenAI's DevDay shipped more than twenty announcements. The platform core is agents: always-on Dots, a cheaper GPT-6.1 Sol, ChatGPT Spaces, MCP webhook events, and a Decisions API.

The decision it puts on your desk

Rewrite every agent-in-a-box integration into Dots this quarter, before your renewal pricing shift lands on existing plans. If you build decision-layer modules, migrate the Decisions API inside two sprints, so latency budgets stay under ~150 milliseconds end to end. And cap AI tenants behind an Ultrafast tier now, so throughput bills as a plan feature rather than a bill shock.

OpenAI's DevDay reset the agent tooling baseline by shipping the same shift once in numbers rather than announcements.

The headline pair, from the launch coverage: "GPT-6.1 Sol (near-Astra performance at about a fifth of the token cost, $2/$10 per million, with an Ultrafast tier at roughly 300 tokens a second)" and Dots, the always-on agents.

Dots change what an agent is. They run on their own cloud machines rather than inside a chat window, and they connect to more than 4,000 apps.

ChatGPT Spaces lands as a shared workspace for humans and agents. MCP Events provides real-time webhook subscriptions from MCP servers for developers who require subscriptions rather than scheduled polling.

The Decisions API lands as the shape change. It returns a parsed choice rather than prose in about 150 milliseconds, which puts a decision judgment inside apps that never needed full text.

One number measures the Sol reduction next to Astra: $0.72 per Artificial Analysis index task, against $3.26 for Astra's top setting, per follow-up testing.

The DevDay stack, by developer surface
The DevDay stack, by developer surface

What each change actually prices

Follow the money next to the cost evolution.

Ultrafast tiering sets throughput as a billable option: a build can pay for speed rather than wait. And the $2/$10 rate takes mid-frontier pricing a tier below GPT-6 Astra's $10/$50 launch list.

Migration cost mostly sits in architecture. Applications already send a single prompt out and read a single text string back, and those need to reformat for decision-shaped payloads on the Decisions API.

Under the hood, the Sol cost curve is the live engineering note. Sol matching Astra's performance at a fifth of the tokens means the inference token bill bends downward, at the price tier your agent budget runs on.

So the Decisions API prices routing rather than replacing models: for classification-shaped jobs, a choice at 150ms costs less than prose at any speed.

Ecosystem check: rivals ship parts of this stack.

Meta and Google both announced agent-platform moves in the same month, but OpenAI's Dots carries the platform-first layer: agents running in their own machines, inside a subscription.

Precedent explains the timing too. Astra launched on September 3, and DevDay landed a cheaper tier within a month.

I get the sense builders face the longer migration list here, and my read is that the real cost is on teams who now hold agent-work sponsorship inside OpenAI's platform themselves.

The unresolved edge is the access detail. Dots and Sol quotas sit inside subscription tiers, and OpenAI named no per-seat pricing inside its DevDay post.

What DevDay resets, in cost order
What DevDay resets, in cost order