GPT-6.1 Sol: The Number That Matters
OpenAI says GPT-6.1 Sol matches Astra on coding and computer use at one fifth of Astra's token cost.
If that holds up on real workloads, it is the most consequential announcement from DevDay, and considerably more so than the agent that got the headlines. Availability: the API, ChatGPT Work, and Codex across Plus, Pro, Business, Enterprise and Edu.
"Matches on coding and computer use" is a narrower claim than "matches Astra". Treat it as scoped to those two areas until independent benchmarks exist.
Pro 500 and the Ultrafast Tier
- Pro 500: $500 per month, includes access to the Ultrafast speed tier
- Ultrafast: up to 8x faster generation in Codex - around 300 tokens per second - at up to 6x standard token pricing
- Available on Pro 500 and Enterprise
So OpenAI shipped a model at one fifth the cost and a speed tier at six times the cost on the same day. Those are aimed at completely different buyers: one at anyone whose bill scales with volume, the other at people for whom a developer waiting is more expensive than the tokens.
The honest way to evaluate Ultrafast is to price the wait. At 6x tokens, it pays for itself only if the time saved is worth more than five extra units of spend. For an agent loop a human is watching, that can be true. For a batch job overnight, it is not.
Everything Else From DevDay
- ChatGPT Space: shared workspace for teams and agents, with collaborative Pages and export to PowerPoint and Google Slides. Pro, Business and Enterprise on desktop and web; Plus gets read-only on mobile.
- Decisions API: limited preview, broader release "in coming days".
- Codex Security Cloud: Pro, Business, Enterprise and Edu.
- Agents API: computer use for developers, on Pro 500 and Enterprise.
- Private Intelligence: zero data retention option.
- OpenAI Marketplace: 32 partner integrations for enterprise customers.
Plus dots, the always-on agents, which we cover separately here.
The Week in Model Pricing
This is the second major price move in two days. On 29 September Anthropic released Claude Sonnet 5.5 at unchanged per-token pricing, with cost per task down up to 30% through fewer tokens and batched tool calls, and outscoring Opus 5.5 on Terminal-Bench 4.0. Our breakdown is here.
Two labs, two days, the same strategy from opposite directions: Anthropic made the mid-tier model good enough to displace the flagship, OpenAI made the cheap model match the expensive one on the tasks people actually run.
The flagship-model-for-everything era is ending for cost reasons, not capability ones.
What to Do About It
- Running Astra in production: test GPT-6.1 Sol on your own workload this week. A fifth of the cost is worth an afternoon of evaluation.
- Considering Pro 500: measure how long your team actually waits on generation before paying $500 a month plus a 6x token multiplier for that wait to shorten.
- Comparing across labs: compare cost per completed task, not per token. Both announcements this week turn on token efficiency, and a per-token table will mislead you.
FAQ
How much cheaper is GPT-6.1 Sol than Astra?
One fifth of Astra's token cost, with OpenAI claiming parity on coding and computer use.
What is Pro 500?
A $500 per month ChatGPT tier that includes the Ultrafast speed tier.
How fast is Ultrafast?
Up to 8x faster generation in Codex, around 300 tokens per second, at up to 6x standard token pricing.
Where can I use GPT-6.1 Sol?
The API, ChatGPT Work, and Codex on Plus, Pro, Business, Enterprise and Edu.