OpenAI's update marathon Day 2: free Auto-review, API tiers cut from 5 to 3, Decisions API goes public beta

机器之心 · wechat · 2026-10-07

OpenAI's Tibo committed to 28 consecutive days of updates for Codex/Work users, with a full usage reset as penalty for any missed day. After Day 1's 50% default speed boost for GPT-6 Astra and GPT-6.1 Sol, Day 2 shipped four updates aimed at friction in real agent workflows:

Auto-review now free: the "Approve for me" mode uses a second agent to vet the primary agent's high-risk actions, free for all ChatGPT-auth users without consuming quota — addressing approval fatigue in long-running tasks.
API tiers cut from 5 to 3: Build ($5), Launch ($100), Grow ($500), with the top tier's threshold halved from $1,000 and automatic upgrades.
Meetings plugin: ChatGPT records meetings and saves summaries/todos to Space for ongoing project planning; macOS beta for Pro/Business users.
Decisions API public beta: GPT-6 Luna-powered API for judgment-style calls (Predicates/Choices/Scores), up to 10x faster than Responses API per OpenAI demos (150ms vs 1.6s), targeting routing, tool selection, classification and agent risk checks.

The common thread: stop chasing capability benchmarks and fix the permission, scaling, context and low-latency decision pain points that surface once agents run for hours.

Original post →

OpenAI's update marathon Day 2: free Auto-review, API tiers cut from 5 to 3, Decisions API goes public beta

More from coding & agent