OpenAI DevDay 2026 Recap: What It Means for Your AI Costs

Last updated: October 2026

Key takeaways

• OpenAI's recap lists 20+ announcements, including GPT-6.1 Sol, Dots always-on agents, Ultrafast, a Decisions API, an Agents API with computer use and a Pro 500 tier.
• Always-on agents turn usage into a clock, so cost grows with time, not just with user actions.
• Sign in with ChatGPT lets plan usage flow to 16 partner tools, which can shift who pays for tokens.
• The OpenAI Marketplace lets enterprises apply existing commitments to partner software, which changes how buyers procure.
• Revisit your pricing before the next quarter, not after.

What did OpenAI announce at DevDay 2026?

OpenAI's recap covers 20+ announcements. The ones that matter most for cost: GPT-6.1 Sol, Dots always-on agents, Ultrafast with up to 300 tokens per second, a Decisions API, an Agents API with computer use, plugin extensions, Spaces and Pages, and a new Pro 500 tier. It also introduced an OpenAI Marketplace and Sign in with ChatGPT. Read the original at DevDay 2026 Recap. For the model itself, see our note on GPT-6.1 Sol launch pricing, where the OpenAI price list lives.

How much does GPT-6.1 Sol cost?

Per the launch coverage, GPT-6.1 Sol is $2 input, $0.10 cached input and $10 output per 1M tokens. That is the rate card for most of the agent features below, so every new capability is a multiplier on those three numbers. For a deeper launch checklist, see frontier model launch pricing checklist for founders.

What do always-on agents cost per month?

Dots always-on agents change the shape of usage. A normal feature costs money when a user clicks. An always-on agent costs money every time it wakes up. Say an agent wakes 24 times a day, sends 10k input tokens and returns 500 output tokens each time, at Sol rates. These numbers are illustrative.

| Scenario (illustrative) | Cost per wake | Wakes per month | Monthly cost per agent | |---|---|---|---| | No caching | $0.025 | 720 | $18.00 | | 8k of the 10k input cached | $0.0098 | 720 | $7.06 |

Per wake with no caching: 10k input tokens at $2 per 1M is $0.02, plus 500 output tokens at $10 per 1M is $0.005, so $0.025. With caching: 8k cached at $0.10 is $0.0008, 2k uncached at $2 is $0.004, plus $0.005 output gives $0.0098. Now multiply by your customer count. A thousand agents is $7k to $18k a month, whether or not anyone is looking at them.

Who pays when users sign in with ChatGPT?

Sign in with ChatGPT lets a user's plan usage flow to 16 partner tools. If you are one of them, part of your token cost can move onto the user's own plan instead of your bill. That can improve your gross margin, but it also gives OpenAI a say in the relationship and can limit your control over which model runs. Treat it as a channel decision, not only a cost decision.

What does the Marketplace change for B2B pricing?

The OpenAI Marketplace lets enterprises apply existing OpenAI commitments to partner software. For a vendor, that can shorten procurement because the buyer already has a budget line. It can also push you toward pricing that maps cleanly to a committed spend. Check how your model rates compare against alternatives on the OpenAI vs Anthropic comparison before you list.

What should founders do this week?

1. 1Re-run your cost forecast at the new Sol rates.
2. 2List every feature that could become always-on and price it per agent or per hour.
3. 3Decide whether Sign in with ChatGPT helps or hurts your margin and control.
4. 4Cap background wakes by default, and let customers opt in to more.

Also see what 14x faster inference does to your cost model, since faster output can also mean more tokens consumed per hour.

Takeaway: DevDay turned several one-off costs into continuous ones, so price time as well as usage.

To re-run your forecast, try the LLM cost calculator.