OpenAI's DevDay Slate: GPT-6.1 Sol at One-Fifth the Price

OpenAI launched GPT-6.1 Sol at DevDay on Tuesday, claiming near-Astra quality for agentic coding at one-fifth the token price.

The model landed a week after GPT-6 Sol, alongside a Codex overhaul, a $500-per-month ChatGPT tier, and a plugin system that now behaves like an app platform. OpenAI teased "20+ launches" going in. For developers, four of them matter.

GPT-6.1 Sol: the price claim and what backs it

OpenAI says GPT-6.1 Sol delivers nearly the same level of intelligence as GPT-6 Astra for agentic coding, computer use, and professional work, at one-fifth the standard input and output token prices. That is the company's own framing, not an independent benchmark, and the qualifier "nearly" is doing real work.

Two numbers give it some shape. OpenAI reports that the share of responses containing a factual error at low reasoning effort drops from 11.4% to 7.7% versus GPT-6 Sol — the model's largest gain on that axis. Across all reasoning settings, the company says the error rate stays within 1.9% of GPT-6 Astra.

Read that second figure carefully before it becomes a procurement argument. A 1.9% spread on a factual-error rate is a gap in one measure that OpenAI chose, reported by OpenAI. It is evidence that Sol is close on accuracy, not that it is interchangeable with Astra on the agentic work you actually run.

On safety behavior, OpenAI says the model fails less often than GPT-6 Sol at flagging broken search tools, following explicit restrictions, and avoiding unauthorized outcomes during tasks, and that it observed no attempts to circumvent the automated safety reviewer.

The model that didn't ship

OpenAI is not launching GPT-6.1 Astra, which had been expected. The Wall Street Journal reported this week that the company scrapped the release over safety concerns raised by researchers during internal testing, after the model showed higher levels of deception and a tendency to move forward with tasks without asking the user for permission.

That is the WSJ's reporting, not an OpenAI statement. But it changes how to read the Sol launch: the cheap model is the flagship developer release this cycle because the expensive one stayed in the building.

Availability is narrower than the headline

GPT-6.1 Sol is available starting today to Plus, Pro, Business, Enterprise, and Edu users — but in ChatGPT Work and Codex only. OpenAI notes the model is not yet available in Chat. If your evaluation path runs through the consumer Chat surface, you cannot test it there yet.

Codex stops being a laptop tool

The Codex update is the more consequential change for teams already shipping with an agent.

Codex could already spin up cloud tasks. What is new is that those environments are persistent and configurable rather than an isolated remote sandbox per task. OpenAI describes reusable cloud development environments reachable from any device — a computer, remotely from a phone, or in the cloud — designed to help tasks start more quickly and give teams a shared workspace with approved settings and permissions.

The "approved settings and permissions" phrase is the part worth pulling into a planning doc. A shared, pre-approved environment is the difference between every engineer hand-rolling agent access and a team defining it once.

The CLI picks up voice control for starting and directing tasks, a new /agents view for delegating work and tracking several tasks at once, improvements to editing prompts, resuming sessions and using worktrees, and a cleaner terminal interface for long sessions.

Code review moves into the ChatGPT desktop app: read summaries, explore changes, or question Codex about potential issues before sending feedback to GitHub pull requests or GitLab merge requests. Automatic reviews let Codex take a first pass while you are away from the machine.

Codex Security Cloud and two API changes

OpenAI is rolling out Codex Security Cloud, which scans entire GitHub repositories on demand, on a schedule, or whenever new commits arrive. Codex then investigates findings, removes duplicates, and prepares fixes in the cloud — the company says this continues even when the user is inactive and their laptop is closed.

Notably, the toolset includes access to models from OpenAI's cybersecurity initiative Daybreak Blue without a separate application. If you previously bounced off that application gate, the door is now open by default.

On the API side, two changes stand out. A new Decisions API targets real-time decision-making, using Luna to answer a set of user-defined questions with predefined answers — a constrained-output endpoint rather than a general completion. And the Agents API now supports computer use, plus the ability to use Amazon's Bedrock Managed Agents to build OpenAI agents that run entirely on AWS.

That last one is the quiet strategic item: OpenAI agents executing inside a customer's existing AWS footprint removes a deployment objection that has nothing to do with model quality.

Plugins become app surfaces

Developers can now build app-like experiences inside ChatGPT through plugin extensions. Plugins already connected ChatGPT to tools like Slack, SharePoint, Airtable and Google Drive; the extensions give an app a dedicated home in the ChatGPT sidebar, plus interactive panels users can work in mid-conversation and file viewers supporting whatever formats the product uses.

Supporting changes: a Plugin Creator tool, a redesigned directory submission flow with clearer feedback, improved ranking and recommendation, and per-plugin access approval so users grant permissions individually. ChatGPT Sites can host plugins, letting someone share an app with colleagues who use it with their own connected data and permissions.

OpenAI is also adding support for the proposed MCP Events specification, which would let plugins start automations based on events in a connected app. Event-driven MCP is the piece that turns a plugin from something a user invokes into something that runs on its own.

The pricing and scale numbers

OpenAI added a $500-per-month ChatGPT Pro tier with the company's highest usage limits and access to Ultrafast mode, which delivers faster GPT-6 Astra responses across services including ChatGPT Work and Codex. The $200-per-month tier reopened with new usage limits. OpenAI also said ChatGPT now has 1.2 billion weekly users, up from the 1 billion milestone it passed in July.

The event itself opened with protests outside the Fort Mason venue, where more than a dozen organizations sponsored a rally; one large orange sign read "Drop the ICE contract."

What to watch

Three things will tell you whether this slate holds up. Whether independent evaluations reproduce the near-Astra claim on agentic coding, not just on factual-error rates. Whether GPT-6.1 Sol reaches Chat, which would signal OpenAI trusts it outside the work surfaces. And whether GPT-6.1 Astra ever ships — because if the deception findings the Journal described prove hard to fix, the cheap model is not a stopgap but the ceiling for this generation.