OpenAI API: GPT-6.1 Sol pricing, Decisions beta, three usage tiers

If you build on OpenAI's API or use Codex, these are the changes since late September that affect cost and speed: GPT-6.1 Sol, a Decisions API beta, three usage tiers, and Ultrafast mode.

What changed#

Sep 29: GPT-6.1 Sol (gpt-6.1-sol) at $2 input, $0.10 cached input, $2.50 cache write and $10 output per 1M tokens (prompts up to 272K), with Multi-agent in beta. GPT-6 Sol lists $0.20 cached input at the same $2/$10. In Codex and ChatGPT Work, OpenAI is rolling it out "starting with Pro users and expanding to Plus, Business, Enterprise, and Edu."
Oct 6: Decisions API in beta with gpt-6-luna, to "Turn text and images into typed answers 10x faster than the Responses API."
Oct 6: Usage tiers simplified "from five to three: Build, Launch, and Grow." Orgs upgrade automatically as credit purchases reach tier minimums.
Oct 8: Ultrafast for gpt-6.1-sol with service_tier: "ultrafast", global processing plus US and EU data residency.
Oct 5: In-product HIPAA BAA acceptance for eligible orgs.
Sep 29: Computer use added to the Agents API.

Why it matters#

Classification and extraction jobs that only need a typed answer could move to Decisions for speed. Tier changes may shift your rate limits, so check them.

How to use it#

In Codex, run /model and pick GPT-6.1 Sol if it appears for your plan.
Check your tier and per-model limits in the usage tiers page.
For latency-sensitive calls, add the Ultrafast service tier:

{"model":"gpt-6.1-sol","service_tier":"ultrafast","input":"Summarize this incident log in 3 bullets."}

Good to know#

The Decisions API is in beta. The 10x speed figure is OpenAI's claim.
Astra Ultrafast does not support EU residency; 6.1 Sol Ultrafast does.
Since Sep 2, fast traffic ramps can return 429 slow_down; honor Retry-After.

Sources#

https://developers.openai.com/api/docs/changelog (checked 2026-10-12 10:43 ET)
https://help.openai.com/en/articles/6825453-chatgpt-release-notes (checked 2026-10-12 10:43 ET)

Last checked: 2026-10-12