ChatGPT is gaining Computer History, a feature that turns activity on your Mac into reusable work context. OpenAI is also testing an especially fast mode for GPT-5.6 Sol and improving ChatGPT’s voice capabilities. If you use Artificial Intelligence (AI) at work or at home, these changes promise less repetition and shorter waits—but only under specific conditions.
What is changing in ChatGPT?
The central addition is Computer History for ChatGPT. It records interactions across apps and websites, converts them into text summaries, and organizes them in a searchable timeline. ChatGPT can later use that information as context, reducing the need for you to explain every previous step again.
Computer History replaces Chronicle, an earlier research preview that relied on screenshots. According to OpenAI, the new system instead captures interaction events such as clicks, keystrokes, keyboard shortcuts, and app changes. It does so through the accessibility system, the part of macOS that lets applications access user interactions.
OpenAI is also launching a preview of Ultrafast for GPT-5.6 Sol. The mode is reported to deliver as many as 750 output tokens per second and to run up to 14 times faster. Tokens are small units of text used to measure processing and, in many services, calculate usage charges.
For a typical ChatGPT user, this speed announcement is less immediate than it may sound. Ultrafast is initially available only to selected customers through the application programming interface (API), which companies use to connect AI models to their own services. OpenAI says access will expand as capacity grows, but the supplied information gives neither a general release date nor an Ultrafast price.
How does Computer History work?
The recorded events are periodically converted into text summaries. A timeline groups them by date and time and shows which apps were involved. The resulting Memory files—stored information that can inform later conversations—are kept locally as Markdown files, a simple readable text format.
One practical example is finding a document you edited recently. Instead of reconstructing its filename and location yourself, you could ask ChatGPT to identify it from your recorded activity. Another example is preparing a daily standup summary, meaning a short team update on work in progress: ChatGPT can use the captured workflow to determine what you worked on.
The system is also intended to recognize recurring workflows. When it detects a repeatable pattern, it can suggest a “skill,” meaning a reusable automation based on that process. This shifts ChatGPT from a tool that receives isolated instructions toward an assistant that already knows parts of your work context.
OpenAI has added several approval layers. In Business and Enterprise workspaces, an administrator must first enable the feature, after which each person must opt in individually. The Memories feature must also be active. Include and exclude lists let you select which apps and websites contribute data, while a macOS menu bar control can pause and resume recording.
According to OpenAI, Computer History does not capture screenshots, screen recordings, microphone input, or system audio. Activity in private browsing mode is also excluded. Those restrictions narrow the recording, but they do not remove every privacy concern: clicks, typed input, and app changes can still produce a detailed account of your working day.
Who benefits from more speed and voice?
Ultrafast initially targets organizations that need an answer while an event is still unfolding. OpenAI’s examples include analyzing logs and changes during a system outage, detecting suspicious financial transactions, and handling customer support. In online retail, an integrated assistant could answer product questions and check inventory before a hesitant shopper abandons a purchase.
Speed itself is becoming a paid service tier. The existing Fast Mode for GPT-5.6 Sol reportedly offers up to 2.5 times the speed and lower latency, meaning less delay, than standard processing, at roughly twice the price. The report expects Ultrafast to cost more again, but no confirmed price is available.
At the same time, prices are falling for other models. A report on pricing pressure in the AI market describes an 80 percent cut for GPT-5.6 Luna: from $1 to $0.20 per million input tokens and from $6 to $1.20 per million output tokens. It attributes the pressure to lower-cost Chinese alternatives and rising AI bills for businesses.
This does not directly contradict the premium charged for greater speed. OpenAI can offer cheaper models for cost-sensitive tasks while charging more when exceptionally fast processing has business value. For organizations, choosing the appropriate model therefore matters more than automatically selecting the most powerful option.
OpenAI presents GPT-5.6 itself as more efficient. In its guide to the GPT-5.6 model family, the company says Sol at low reasoning effort outperformed GPT-5.5 at high reasoning effort in one evaluation. Such provider claims and reports from participating startups are useful context, but they are not equivalent to independent verification.
ChatGPT Voice is also meant to feel more immediate. GPT-Live-1, released in July 2026, is designed to sound more human and work as a live translator. In a test involving a French language-learning podcast, the German translation was accurate in substance, but the source and translation played at the same time. That made listening difficult even after the original audio was turned down.
Pros and Cons of the new ChatGPT features
Pros:
- Less repetition – Computer History can supply previously edited documents and earlier work steps as context.
- Practical automation – Recurring workflows can be recognized and potentially turned into reusable templates.
- Faster responses – Ultrafast can sharply reduce waiting during time-sensitive work when access is available.
- More natural interaction – Voice can make conversations and translation more accessible when typing is inconvenient.
Cons:
- Unencrypted memories – Memory files remain as plain text on the Mac until you delete them manually.
- Limited availability – Computer History is not currently offered in Switzerland, while Ultrafast is restricted to selected API customers.
- Unclear additional costs – Fast Mode already carries a substantial premium, and an Ultrafast price has not been provided.
- Errors behind confident delivery – Voice translations can still hallucinate, producing plausible but incorrect information.
The local storage used by Computer History deserves particular attention. OpenAI says temporary event files are deleted after 48 hours, but the Memory files created from them are not removed automatically. Because those files are unencrypted in the file system, other programs running under the same macOS user account could potentially access them.
A correct translation in one live test does not guarantee that later conversations will be accurate. The voice feature can sound fluent and persuasive even when an individual statement is wrong. That may be merely irritating with casual material, but it could contribute to poor decisions in business discussions or other important situations.
What does this mean for your daily work?
If you are a beginner, first check which feature is actually available to you. You cannot currently use Computer History in Switzerland, and Ultrafast is not a generally available switch inside ChatGPT. Voice can instead be tested with a low-risk translation or conversation in which you can compare the output with the original.
If Computer History becomes available later, a limited trial makes more sense than immediately opening your entire working day to it. You can include only selected apps and websites, pause the recording when needed, and review the stored files. Excluding confidential work areas reduces the context being collected, although responsibility for those choices still rests with you and, where relevant, your organization.
Advanced users and organizations can get more value by matching the feature to the task. Recurring document searches and standup summaries are natural candidates for Computer History, while time-sensitive customer support may justify Fast or Ultrafast processing. Large volumes of less urgent work may be better suited to a cheaper model, particularly as the sources describe increasing pricing pressure and usage-based billing.
For Swiss organizations, availability is currently the clearest constraint: Computer History is not offered in the European Economic Area, Switzerland, or the United Kingdom. Layered consent and app filters are sensible controls, but they do not answer when the feature will arrive in Switzerland. Local plain-text files would also remain a security consideration even when recording had been enabled deliberately.
The three developments address different problems: Computer History reduces missing context, Ultrafast shortens response times, and Voice makes spoken interaction easier. Their benefits are tangible, but they are not equally available, free of added costs, or reliably accurate. The largest unresolved risk is that a more informed and convincing assistant may both collect more sensitive context and present incorrect results with greater credibility.
Sources
- ChatGPTs neue “Computer History” verwandelt den Arbeitsalltag in KI-Kontext – The Decoder, 2026-08-14
- 14x schneller: OpenAI launcht neuen “Ultrafast”-Modus für GPT-5.6 Sol – The Decoder, 2026-08-14
- ChatGPT Voice Live als Übersetzer im Test: Warum der neue “Assistent” nervt – t3n, 2026-08-13
- OpenAI and Anthropic in price war as Chinese AI rivals gain ground – Ars Technica, 2026-08-14
- The builder’s guide to GPT-5.6 – OpenAI, 2026-08-13
