OpenAI's GPT-6.1 Sol Cuts Frontier Pricing 80 Percent: What DevDay 2026 Actually Changed

OpenAI's DevDay 2026 keynote landed on September 29 with six announcements bundled into one presentation rather than a single headline release, and that bundling is worth untangling before deciding which part of it matters for your stack. The centerpiece is GPT-6.1 Sol, which OpenAI describes as delivering near Astra level intelligence at a fifth of the standard token price. The rest of the announcement, an Ultrafast inference tier, always on Dots agents, a new Pro 500 subscription, and a Decisions API for real time agentic workflows, are separate products that happened to ship on the same day.

What shipped, in one table

Announcement	What it is	Who gets it
GPT-6.1 Sol	Upgraded Sol tier model, priced at roughly a fifth of GPT-6 Sol's standard rate	API, Plus, Pro, Business, Enterprise, Edu
Ultrafast tier	Up to 8x faster generation in Codex, 6x faster in the API	Pro 500 and Enterprise plans
Dots agents	Always on agents that learn user priorities and handle ongoing tasks	Pro and Business Premium, eligible markets
Pro 500 plan	New subscription at 25x the ChatGPT Plus usage allowance, includes Ultrafast	New tier, priced above existing Pro
Decisions API	Real time decision making against predefined answer sets	API, general availability with the DevDay release
Private Intelligence	Zero Data Retention with Private Safety Processing, Private Inference coming	API and enterprise tiers

The pricing math OpenAI wants you to do quickly

OpenAI's own framing for GPT-6.1 Sol is near Astra intelligence at a fifth of the standard input and output token price. Taken against GPT-6 Sol's previously reported pricing of $2 per million input tokens and $10 per million output tokens, a fifth works out to roughly $0.40 input and $2.00 output per million tokens. That is a genuinely large cut, not a rounding adjustment, and it lands GPT-6.1 Sol close to where the lower end of the frontier tier was pricing itself only weeks ago.

The claim worth slowing down on is "near Astra intelligence." Astra is OpenAI's top tier model, and if GPT-6.1 Sol is delivering comparable output quality at a fifth of Sol's own prior price, that is a different statement than delivering comparable quality at a fifth of Astra's price, which would be a far larger claim. OpenAI's own wording ties the discount to Sol's existing price point, not to Astra's, so the honest read is that this is Sol getting meaningfully better and meaningfully cheaper at the same time, not Astra's capability becoming available for a fifth of what Astra itself costs.

Ultrafast and Pro 500: speed as a subscription feature

The Ultrafast tier is a separate lever from GPT-6.1 Sol's pricing. It delivers up to 8x faster token generation in Codex and 6x faster through the API, and it is gated behind the new Pro 500 plan or an Enterprise agreement rather than bundled into the standard API price. Pro 500 itself is framed around usage volume, 25 times the ChatGPT Plus allowance, with Ultrafast access included. For a team doing heavy Codex based development work, the relevant comparison is not GPT-6.1 Sol's token price in isolation, it is whether the throughput gain from Ultrafast changes the economics of an agentic coding workflow enough to justify the Pro 500 tier on top of API spend.

Dots agents: always on, not just longer running

Dots agents are described by OpenAI as always on agents that learn a user's priorities and handle ongoing responsibilities, available on Pro and Business Premium in eligible markets. This is a different product category from a long running agentic task inside Codex or the Agents API. A Dots agent persists across sessions and is framed around ongoing responsibility rather than task completion, which raises the same governance questions Contra Collective has flagged in AI agent rollouts generally: who owns an always on agent's actions when nobody is actively supervising a given session, and what the audit trail looks like when the agent's task list changes on its own initiative rather than by explicit instruction each time.

The quieter announcements: Decisions API and Private Intelligence

The Decisions API is a real time decision making capability built around predefined answer sets, aimed at scenarios where an agent needs to make a bounded choice quickly rather than reason open endedly. It is a narrower, more deterministic tool than a general agent loop, and it is the kind of primitive that fits naturally into commerce and workflow automation where a small number of valid outcomes exist and speed matters more than exploratory reasoning. Private Intelligence pairs Zero Data Retention with Private Safety Processing now, with Private Inference described as coming, which signals OpenAI anticipating enterprise procurement conversations where data handling terms are the actual blocker rather than model capability.

What to verify before committing budget

Three things are worth confirming directly against OpenAI's own documentation rather than the keynote framing before making a purchasing decision: the exact GPT-6.1 Sol pricing once it posts to the official pricing page, since the fifth of standard framing is a multiplier on a baseline that may shift by the time general availability lands; whether your actual workload benefits from Ultrafast's throughput gain enough to justify Pro 500 over standard API access, which depends heavily on whether your bottleneck is generation speed or something else in your pipeline; and whether a Dots agent's always on model fits your team's existing agent governance approach, since an agent that acts without an explicit per session trigger needs a different oversight pattern than one invoked on demand.

When this applies to your stack

If your team is already running meaningful spend against GPT-6 Sol or evaluating Codex for agentic coding work, the GPT-6.1 Sol pricing cut and the Ultrafast tier are worth a real cost model rather than a reaction to the keynote framing. If you are evaluating always on agents for the first time, Dots agents raise the same governance questions as any persistent agent deployment, regardless of which vendor ships it.

Contra Collective builds and cost models production AI infrastructure for enterprise clients, including the vendor pricing analysis and agent governance frameworks that turn an announcement like this into an actual budget and rollout plan rather than a reaction to a keynote slide. If your team is evaluating a move to GPT-6.1 Sol or a Dots style always on agent, that analysis is worth doing before commitments are made, not after.

FAQ

Is GPT-6.1 Sol actually a fifth of Astra's price? No. OpenAI's own framing ties the fifth of standard pricing discount to Sol's prior price point, not Astra's. Calculated against GPT-6 Sol's previously reported $2 input and $10 output per million token pricing, GPT-6.1 Sol lands near $0.40 input and $2.00 output, a real cut but not Astra level pricing.

Do I need the Pro 500 plan to use GPT-6.1 Sol? No. GPT-6.1 Sol itself is available across API, Plus, Pro, Business, Enterprise, and Edu tiers. Pro 500 specifically gates the Ultrafast speed tier, which is a separate feature from the model itself.

What is different about Dots agents compared to a normal agentic task? Dots agents are described as always on and persistent across sessions, learning user priorities and handling ongoing responsibilities, rather than being invoked for a single bounded task the way a Codex agent run or an Agents API call typically is.

Should enterprise teams wait for Private Inference before adopting GPT-6.1 Sol? That depends entirely on your data handling requirements. Zero Data Retention with Private Safety Processing is available now; Private Inference is described as coming. Teams with strict data residency requirements should confirm current Private Intelligence coverage against their specific compliance needs rather than assuming Private Inference timing.