OpenAI AI Updates: October 15, 2026

1. GPT-5.5 Retires From Codex for ChatGPT-Authenticated Users

OpenAI. As of October 14, 2026 at noon PDT (19:00 UTC), GPT-5.5 is no longer available in Codex for users signed in with a ChatGPT account. The OpenAI API and Codex sessions authenticated with an API key are not affected. OpenAI recommends switching to GPT-6.1 Sol (gpt-6.1-sol) where the plan and client support it, and tells teams to update workspace defaults, saved model settings, managed configurations, custom agents, scheduled tasks, and scripts that still select gpt-5.5. Source

2. OpenAI Economic Research Reports Agentic Tools Now Produce Over Half of EU Enterprise Output Tokens

OpenAI. A new OpenAI Economic Research report finds that Codex and ChatGPT Work account for more than half of output tokens among ChatGPT Enterprise customers in its EU sample, with the share ranging from 22% in Slovenia to 85% in Lithuania. Weekly active users of those agentic tools at EU businesses with 200 or fewer users grew 8.1x between May and early October 2026. Firms in the top 10% by output tokens per active user in the EU generated more than four times the tokens of typical EU firms in September, but frontier firms outside the EU generated more than twice as many as EU frontier firms, and EU companies make up only 10% of the global frontier group. The report comes with an interactive dataset and PDF. Source

3. Third Moment Research Runs 10,000 Concurrent Codex Agents on 36 to 48 Hour Sessions

OpenAI. An OpenAI case study describes how Third Moment Research runs 10,000 agents concurrently on autonomous quantitative finance research sessions lasting 36 to 48 hours, at one point consuming 1 trillion OpenAI tokens in a single day. Agents run the default Codex harness in Docker with a terabyte of data, and Codex is wrapped as a subagent inside an outer GPT-6 Sol or GPT-6.1 Sol harness that inspects unfinished trajectories and decides whether to retry, terminate, or escalate to a human. Codex profiled the team’s backtester and added caching and memory packing that cut a year-long order-level backtest from about 3 hours to 3 to 5 minutes, reducing token waste from agents polling for completion. The team names GPT-6.1 Sol its most-used model for token efficiency and uses OpenAI’s Daybreak platform to have agents penetration-test the sandboxes before simulations launch. Source