One change today, two useful Microsoft catch-ups

OpenAI increased the default GPT-6.1 Sol context length in Codex and Work on October 12. For people carrying a complex task across many files and decisions, the practical change is more room for relevant material before a conversation needs to be compressed. This brief also revisits two separate Microsoft announcements from October 7 and 9. Their dates matter: the Windows and decision-model developments are catch-up items, not launches on October 12.

GPT-6.1 Sol gets a 500,000-token default in Codex and Work

OpenAI's Tibo announced that GPT-6.1 Sol now starts with a 500,000-token context in Codex and Work, up from 272,000. Users can override the default, with a maximum of one million tokens. OpenAI expects the larger default to reduce compactions during longer tasks. That is a change to how much context the model can hold at once, not a promise that it will recall every detail perfectly or finish a task without review.

A practical use is a repository investigation that needs a bug report, several relevant files, test output and earlier decisions in the same working context. With more space, the assistant may need to summarize and discard earlier detail less often. You should still point it to the specific failing case and inspect its final change. A larger window helps preserve evidence; it does not establish that the assistant actually read the right files, ran the tests or respected every constraint.

Tibo also said OpenAI is not charging long-context fees for GPT-6.1 Sol usage drawn from a subscription. That statement does not say the subscription has unlimited usage or that other billing routes have identical terms. If a team has manually set a smaller context limit, the new default may not be the value it is using; check the active configuration before drawing conclusions about a particular session.

Windows gives agents a boundary before broader local Copilot arrives

On October 7, Microsoft set out Windows for hybrid intelligence: agents can use local models for suitable work and reach cloud models when needed. The concrete availability change is Microsoft Execution Containers, or MXC, which Microsoft says is generally available on Windows 11. Developers and administrators can set file and network limits outside an agent's control, then run its workload inside an enforced boundary. Separate agent identity and wider governance integrations are part of Microsoft's direction, with some components still forthcoming.

For a coding agent, a useful policy might allow edits in one project folder and access to its build tools while blocking personal documents and production settings. The boundary is useful only if someone defines and checks it against the agent's real work. Microsoft's developer guidance describes different container options and trade-offs, so MXC availability alone does not mean every existing agent already runs with the appropriate permissions.

Other October 7 features have later access dates. Microsoft says GitHub HydraFusion routing to local models is due in experimental preview later in October in the Copilot app, CLI and VS Code. Copilot's permission-based access to PC context, local actions and local models is expected to begin rolling out on Copilot+ PCs over coming months. Availability can vary by device, market and silicon; those features should not be treated as present on every Windows PC today.

Microsoft-Decision-1 chooses among defined options

Microsoft introduced Decision-1 separately on October 9. It returns a probability score for supplied choices, such as whether to continue a workflow, retry a step, route a request or ask a person to review. Developers can access it through Microsoft Foundry or OpenRouter. Unlike a general chat model, its purpose is the repeated, constrained decisions inside software and agent workflows, where each call's delay and cost add up.

Consider an intake flow with three permitted outcomes: answer from an approved knowledge base, request missing information or send the case to a human. A small decision call could rank those options, while the application applies its own threshold and policy before taking action. That is an editorial example, not a tested XANTRANS deployment. The model's score should be checked against representative cases before it is trusted to route consequential work.

Microsoft reports roughly 35 times lower median latency than GPT-6 Sol in its structured decision benchmark. This is a provider measurement for that task type, not a general claim that Decision-1 is better or faster at writing, coding or open-ended reasoning. The announced price is $0.042 per million input tokens with no charge for output tokens; deployment costs and terms should be checked in the chosen service.

Sources & verification

Prepared with AI assistance from the linked official sources. Practical examples are editorial possibilities; XANTRANS has not independently tested these features or benchmarked the models.