Week ahead: Microsoft reports, OpenAI Evals goes read-only, Mistral's weights deadline arrives

Three dated items fall between Sunday, October 25 and Saturday, October 31. Microsoft reports quarterly results on Wednesday. OpenAI's Evals platform stops accepting changes on Saturday. And Mistral's own deadline for releasing the Mistral Large 4 weights runs out at the end of the month. Everything below comes from a date the company itself has published. We did not include dates that only appear in third-party earnings calendars.

Wednesday, October 28: Microsoft reports

Microsoft said on October 7 that it will publish results for the first quarter of fiscal 2027 (July to September) after the market closes on Wednesday, October 28. The webcast starts at 2:30 p.m. Pacific Time, which is 18:30 in Buenos Aires.

For AI readers, this call is one of the few regular times Microsoft gives numbers on Azure demand, capital spending and data-center capacity. Those numbers show how much compute is being built for the models people use through Azure and Copilot. We are not previewing figures: Microsoft has not said what it will disclose, and analyst estimates are not reporting. If something material comes out about capacity, pricing or AI infrastructure, the daily edition will cover it.

We could not open other large AI companies' official earnings-date announcements from this environment, so no other calls are listed. We would rather leave them out than repeat dates we could not check.

Saturday, October 31: OpenAI Evals becomes read-only

On June 3, OpenAI said it is deprecating its Evals platform. Its deprecations page sets two dates. On October 31, 2026 existing evals become read-only. On November 30, 2026 the Evals dashboard and API are scheduled to shut down. Reusable prompt objects (the v1/prompts API) and Agent Builder are also scheduled to shut down on November 30. They were announced the same day, but they do not have an October step.

In practice, Saturday is the last day to edit an eval or add test cases on OpenAI's platform. After that, teams have a month to move what they need. OpenAI points developers to Promptfoo, an open-source evaluation and red-teaming tool. Its migration guide says the move is manual: you recreate each eval's prompts, providers, test cases and assertions in a config file, and the process does not rely on an Evals export feature. The guide also notes that runs in Promptfoo are new runs, separate from any results already in OpenAI's dashboard.

If your model-selection decisions depend on regression suites stored in OpenAI Evals, this is the item that needs action this week. Write down the test data and grading criteria while you can still see them, and keep a copy of past results if you want a baseline to compare against when you move to a new model.

By October 31: Mistral's deadline for Large 4 weights

When Mistral launched a public API preview of Mistral Large 4 on October 6, it said the weights would come out "by the end of the month", along with more detail on the architecture, more benchmarks and its post-training method. Mistral describes the model as a natively multimodal mixture-of-experts with 1 trillion total and 52 billion active parameters. Mistral lists it at $1.36 per million input tokens and $4.18 per million output tokens. Until the release, Mistral says, it is red-teaming the model with cybersecurity leaders, vetted partners and state authorities.

Mistral has promised a month, not a specific day. As of Sunday evening, no Large 4 repository appeared on Mistral's Hugging Face page. The launch post does not name a license for the weights, and it does not say whether the released weights will match the preview. That matters because Mistral says the reinforcement-learning run behind the preview is "still in flight" and expects "large and rapid improvements", so what ships may differ from the API version people have tested. The benchmark figures in the launch post, including cybersecurity results that compare Large 4 with closed models that refused the task, are Mistral's own testing plus one third-party index it cites. They will be easier to check once the weights are public.

If the weights miss the end of October, that is also news. We will report it either way.

What we left out

Conference dates we could not confirm on official pages are not listed. We also left out unannounced model launches and rumored dates.