Week ahead: CFTC's compute-futures comments close, OpenAI retires gpt-4 and o1

Three dates between Sunday, October 18 and Saturday, October 24 are confirmed in primary documents, and two of them are developer deadlines. A U.S. regulator closes comments on whether AI compute can be traded like a commodity, IBM reports third-quarter results, and OpenAI and Google switch off a long list of older models and previews.

Tuesday, October 20: CFTC comments close on compute derivatives

The Commodity Futures Trading Commission published a request for comment on August 21 about futures and other derivatives that settle to the price of computing power. The Federal Register record says comments must be received on or before October 20, 2026, under docket CFTC-2026-1850, and the docket record it mirrors from regulations.gov shows late comments are not accepted. By Sunday it showed 52 comments.

The notice is a request for input, not a proposed rule with contract terms. It is specific about what the CFTC thinks is hard. It describes compute markets as fragmented, with prices set mostly in private bilateral deals, and it warns that dominant providers could hold enough pricing power to make a settlement index manipulable. The agency says it preliminarily believes compute may not yet be fungible, standardized or liquid enough to back a typical commodity contract.

The part most relevant to Promptea readers is what could be the underlying. The CFTC's example is the hourly rental price of a B200 GPU, but it also names "access to a stated volume of LLM inference tokens" as a possible underlier. It asks how much compute trades at published prices, whether providers who post rates could move an index by changing them, and whether an exchange could surveil a market built mostly on private contracts. The notice cites the White House's July 2025 AI Action Plan, which recommended improving the financial market for compute so startups and academics can get large-scale capacity.

Nothing is decided on Tuesday. The deadline only closes the record, and the notice gives no timeline for what the CFTC does next.

Wednesday, October 21: IBM reports third-quarter results

IBM said on October 7 that it will discuss third-quarter results on a call at 5:00 p.m. ET on Wednesday, October 21 (18:00 in Buenos Aires), with the earnings release posted to its investor site before the webcast.

The bar comes from IBM's second-quarter release on July 22. The company said it faced revenue headwinds late in that quarter and now expected full-year constant-currency revenue growth of four to five percent, with free cash flow up about $1 billion year over year. It named Red Hat, watsonx, HashiCorp and Confluent as the software lines serving AI demand. Those are IBM's own expectations; Wednesday is when they meet reported numbers.

Thursday and Friday, October 22–23: model shutdowns at Google and OpenAI

Google's Gemini API deprecations page lists October 22 as the shutdown date for veo-3.1-generate-preview, veo-3.1-fast-generate-preview, veo-3.1-lite-generate-preview and gemini-omni-flash-preview, all pointing to gemini-omni-1.1-flash as the replacement. The deep-research-pro-preview-12-2025 agent is listed for October 23, replaced by deep-research-preview-04-2026. Google describes these as the earliest possible shutdown dates.

OpenAI's deprecations page lists October 23 for the legacy snapshots it announced in April:

• gpt-3.5-turbo and gpt-3.5-turbo-0125 → gpt-5.6-terra
• gpt-4, gpt-4-0613, gpt-4-1106-preview, gpt-4-turbo and gpt-4o-2024-05-13 → gpt-5.6-sol
• o1, o1-pro and o3-mini → gpt-5.6-sol (o1-pro maps to reasoning.mode: pro)
• o4-mini → gpt-5.6-terra; gpt-4.1-nano → gpt-5.6-luna
• gpt-image-1 → gpt-image-2.5-sunburst or gpt-image-2.5-flare
• Fine-tuned models on gpt-3.5-turbo, gpt-4, gpt-4.1-nano, o4-mini, babbage-002 and davinci-002

If any of these still sit in production, treat the swap as a migration. A replacement can read the same system prompt differently, format JSON or tool calls differently, spend more reasoning tokens and charge a different price per token. Run your evaluation set and check your parsers against the new model before Friday. Fine-tuned models are harder: the fine-tune itself goes away, so the behavior it encoded has to be rebuilt in prompts or in a new fine-tune where that is still available.

Already dated, further out

On October 14 OpenAI added a second wave to the same page: gpt-4o (2024-08-06 and 2024-11-20 snapshots), gpt-4o-mini, gpt-4.1 and gpt-4.1-mini shut down on January 14, 2027, with gpt-6-luna listed as the replacement. That falls outside this week, but it covers models far more widely used than the October batch.

What is not on this list

PyTorch Conference North America and the co-located AGNTCon + MCPCon are listed for this week in San Jose, but the Linux Foundation and PyTorch sites could not be reached from our research environment, so we could not confirm the dates on a primary page and left them out. As always, we do not list model launches that nobody has dated.