AI STRATEGY

OpenAI DevDay 2026: What Every AI Product Manager Needs to Know

By Institute of AI PM·16 min read·Oct 6, 2026

TL;DR

OpenAI held DevDay 2026 on September 29 in San Francisco and shipped 25 announcements. The three that change how you build products now: GPT-6.1 Sol (near-Astra results at one-fifth the price, 1.05M context), Dots (always-on agents that persist between user sessions), and the Agents API with computer use (your product can now instruct an agent to operate a browser or desktop app). The rest of DevDay is mostly developer experience improvements, a new $500 Pro Plus plan for power users, and distribution plays via Sign In with ChatGPT. This article covers what each announcement actually means, which three to act on in the next 30 days, and the strategic pattern behind all of it.

The AI PM Minute

One tactic to make you a sharper AI PM, twice a week. 60 seconds to read. Free.

No fluff. Unsubscribe anytime.

GPT-6.1 Sol: The Model Release That Changes Your Cost Math

GPT-6.1 Sol is an update to GPT-6 Sol, OpenAI's mid-tier frontier model. The headline numbers: 1.05 million token context window, 128,000 token output limit, $2 per million input tokens and $10 per million output tokens. OpenAI's own benchmarks put it at near-Astra quality on coding and computer use tasks, at one-fifth of Astra's price.

That pricing claim is worth examining carefully. "Near-Astra" on OpenAI's benchmarks is not the same as near-Astra on your specific use case. The gap between Astra and Sol tends to be smallest on structured tasks with clear right answers (code generation, data extraction, tool use) and largest on open-ended reasoning, nuanced judgment, and novel problem-solving. Evaluate on your actual workload, not on MMLU or HumanEval.

If you are currently on GPT-6 Astra for coding features

Test GPT-6.1 Sol on your coding workload immediately. If quality holds, you are looking at an 80% cost reduction. That is not a performance optimization, it is a margin decision. Run an A/B test with a 500-request sample before moving.

If you are currently on GPT-6 Sol Luna for agentic tasks

GPT-6.1 Sol is the better version at the same price tier. The 1.05M context window matters for agentic tasks where you need to pass full conversation history and tool outputs. Migrate and run evals to confirm parity on your task distribution.

If you are currently on a smaller model for cost reasons

The $2 input pricing puts GPT-6.1 Sol in the range where it becomes worth testing for tasks you were previously handling with a cheaper model. Run a quality comparison. You may be able to retire a fine-tuning investment if the base model is now good enough.

If you are building document-heavy features

The 1.05M context window is the real news for document-heavy use cases. You can now pass entire product documentation, support ticket histories, or long-form contracts in a single call. Design around the context window, not around chunking workarounds.

Dots: Always-On Agents and What They Mean for Product Design

Dots is OpenAI's always-on agent product: a persistent AI presence that runs in the background between user sessions, monitors events, and takes action when triggered. Unlike a chatbot that exists only when a user opens a conversation, Dots persists across sessions and can initiate actions proactively based on conditions you define.

The product design implications are significant. Most AI features today are reactive: the user initiates, the AI responds. Dots is a model for proactive AI: the AI monitors a stream of events and surfaces the ones that matter, or takes action automatically when a condition is met. This is a fundamentally different product paradigm than a chatbot or copilot.

What Dots enables for consumer products

Persistent assistants that remember context between sessions and act on a user's behalf without explicit triggering. A Dots-powered finance app could monitor account activity and alert you before a charge causes an overdraft, without waiting for you to ask.

What Dots enables for enterprise products

Workflow monitors that watch for conditions and route work. A Dots-powered CRM could watch a deal's activity signals and escalate to a rep when engagement drops below a threshold, without any rep having to check a dashboard.

The trust and permission design challenge

Always-on agents need a clear permission model. Users need to understand what Dots can and cannot do without explicit instruction, and they need easy controls to review and revoke actions. This is a new UX surface area that most teams have not designed before.

The competitive moat question

If OpenAI's Dots is in your product category, you now compete with an always-on agent backed by a frontier model and deep ChatGPT integration. The question is whether your agent has proprietary data or workflow access that OpenAI does not. If not, your time to differentiate is now.

Related: Full Dots deep dive

The knowledge hub has a dedicated analysis of Dots covering the architecture, permission model, and competitive implications in detail. See ChatGPT Dots for Product Managers.

Codex Cloud and the Agents API: OpenAI's Developer Platform Push

Two announcements that together signal OpenAI's ambition to own the agentic development layer: Codex in the cloud and the Agents API.

Codex Cloud moves OpenAI's coding agent out of the ChatGPT interface and into an API-accessible service. You can now call Codex as a service from your application: pass a codebase and a task description, and Codex returns a pull request. This is not just a ChatGPT feature anymore, it is an infrastructure component you can wire into your CI/CD pipeline, your internal developer tools, or your own product.

The Agents API adds computer use to the agentic capability set. An agent orchestrated through the Agents API can now take control of a web browser or desktop application, navigate interfaces, fill in forms, extract data from web pages, and complete multi-step tasks across multiple applications. This is a substantial expansion of what your agents can do without a custom integration for every target system.

Codex Cloud

Best use for product teams: Automate internal development tasks, accelerate your own product's code-generation features, or build developer tools that issue pull requests programmatically. The most immediate use case is internal: replacing ad-hoc coding agent use with a structured API call that logs inputs and outputs.

Watch out for: Codex output quality still requires review. Building a workflow that ships Codex pull requests without human review is a reliability risk. Design for human-in-the-loop by default until you have enough data to know where Codex fails.

Agents API with computer use

Best use for product teams: Automate tasks that currently require a human to navigate a GUI. Examples: scraping data from a vendor portal that has no API, filling in enterprise software forms, testing your own product's UI in a way that mirrors real user behavior.

Watch out for: Computer use is slower and more expensive than API calls. It is the right tool when no API exists, not a replacement for purpose-built integrations. Build API integrations first and use computer use as the fallback for systems with no API access.

Translate AI Announcements Into Product Decisions

The AI PM Masterclass teaches you a framework for evaluating model releases and platform shifts so you can act fast on what matters and ignore the noise. Taught live by a Salesforce Sr. Director PM.

The Rest of DevDay: Ultrafast, the $500 Plan, and Sign In with ChatGPT

Twenty-five announcements is a lot. Here is a triage of the remaining major items ranked by impact on product teams.

Ultrafast speed tier

High

OpenAI launched an Ultrafast speed tier that prioritizes latency over cost, aimed at interactive use cases where users notice response delay. This is the right tier for real-time voice, autocomplete, and any feature where users are watching a cursor blink. Pricing not yet public at time of writing, but expect a premium over standard.

$500 ChatGPT Pro Plus plan

Medium

A new $500 per month tier for power users who need maximum rate limits, early access to new capabilities, and dedicated compute. For AI PMs who rely on ChatGPT for internal work, this is a budget line item worth evaluating if you consistently hit rate limits on the $200 Pro plan. It is not a product API tier.

High for certain categories

OpenAI launched OAuth-style authentication so users can sign into third-party apps with their ChatGPT account and optionally share their subscription plan credits with that app. The plan-sharing permission is the differentiating feature: users can let your app consume their ChatGPT credits, lowering your API cost. See our dedicated analysis for the full strategy read.

ChatGPT collaborative workspace

Low for builders

A shared workspace inside ChatGPT for teams to collaborate on conversations and projects. This is a product feature for ChatGPT users, not a capability you build on. It increases ChatGPT stickiness for enterprise teams, which affects your competitive positioning if you are building an AI assistant for enterprise.

Expanded Realtime API with vision

Medium

The Realtime API (which powers low-latency voice and live audio processing) was expanded to support visual inputs. You can now build real-time multimodal experiences where the model processes a live camera feed alongside audio. The primary use cases are accessibility, live customer support with screen sharing, and field operations tools.

The Strategic Pattern: OpenAI Is Building a Closed Platform

Reading DevDay as a collection of individual product releases misses the pattern. OpenAI is systematically building a closed ecosystem where each component reinforces the others. The pattern becomes clearest when you map the components:

Identity layer

OpenAI controls who users are across third-party apps. This is how Apple built the iOS ecosystem: control identity and you control the relationship.

Presence layer

Dots (always-on agents)

OpenAI is always running in the background, not just when a user opens ChatGPT. This is the ambient computing layer, and it is designed to make ChatGPT as essential as a smartphone's notification system.

Action layer

Agents API with computer use, Codex Cloud

OpenAI agents can now act across any software interface. Combined with Dots, this means an OpenAI agent can be triggered by an event, open an application, complete a task, and report back, all without human initiation.

Data layer

ChatGPT collaborative workspace, plan credit sharing

User activity data flows back to OpenAI regardless of which app captures it. The more products integrate Sign In with ChatGPT, the more behavioral data OpenAI accumulates for model improvement.

The implication for product strategy is straightforward: building on OpenAI's platform is a bet that the platform remains open, benevolent, and competitively priced. Those bets have been right for most of 2023 to 2025. As the platform becomes more tightly integrated with user identity and behavior, the cost of switching increases. Now is the right time to evaluate which parts of your product need to be portable and which can afford the platform dependency.

What to Do in the Next 30 Days

25 announcements creates noise. Here is the prioritized action list for a typical AI product team.

Week 1

Evaluate GPT-6.1 Sol on your highest-cost API workloads

The cost reduction opportunity is immediate. Pick your top three most expensive API calls by volume, run 500 requests through GPT-6.1 Sol, and compare quality scores against your current model. If quality holds, the migration case is trivial. If it does not, document the gap for the next evaluation.

Week 2

Prototype one Agents API use case with computer use

Identify one task in your product or operations that requires navigating a web interface without an available API. Build a minimum viable prototype using the Agents API with computer use. This is your fastest path to understanding the real reliability and latency profile of computer use in your context.

Week 3

Decide on Sign In with ChatGPT

Review our dedicated Sign In with ChatGPT analysis and answer three questions: Does your user base overlap significantly with ChatGPT Pro users? Does the plan credit sharing meaningfully change your unit economics? Does integrating OpenAI identity create a dependency you cannot afford? This decision has a window: early adopters get better placement and integration support.

Week 4

Map your OpenAI platform exposure

List every part of your product that depends on OpenAI APIs, models, or identity. Categorize each as portable (can switch providers in under two weeks), dependent (switching takes one to three months), or locked (switching would require a redesign). Any locked dependency that is not a deliberate strategic choice is a risk item for your roadmap.

Keep Up With AI Without Getting Buried in Noise

The AI PM Masterclass teaches you how to evaluate model releases, platform shifts, and architectural decisions with a framework that transfers across every vendor announcement. Taught live by a former Apple and Salesforce Sr. Director PM.

→ ChatGPT Dots for Product Managers: Always-On Agents and What They Change → Sign In with ChatGPT: What Product Teams Need to Decide → OpenAI's Superapp Bet: Product Strategy Lessons from the ChatGPT Consolidation → GPT-6 Sol Luna for Product Managers: When to Use It and When to Skip It

Before you go: get the AI PM Minute