Last month, we explained how the cost-based usage limits work and which levers get the most out of your allowance. Back then, it was mostly a matter of discipline, because you had to keep the rules in your head without being able to see them in the interface.

That has now changed. Langdock shows your consumption directly in three places, and that does more than it may sound at first. When you can see, while working, what a model costs and how your own chat is filling up, you develop a feel for it over time.

There are three views that, together, paint a full picture:

• The consumption indicator on the model pickershows, before you even start typing, how expensive a model is compared to the others.
• The context indicator in the chatshows live how much of your context window the current chat is currently occupying.
• The session and weekly windowshows how much of your allowance you have already used in the last few hours and this week.

In the model picker, every model now carries a consumption indicator that ranges from low to high. Behind it sits a relative scale that treats GPT-5.2 as the baseline of 1.0. Everything else is measured against that.

The official factors make clear just how big the differences are:

Model	Factor
GPT-5.4 Mini	0.3x
Haiku 4.5	0.8x
GPT-5.2	1.0x (baseline)
GPT-5.4	1.5x
Sonnet 4.6	3.1x
GPT-5.5	about 3.7x
Opus 4.7	8.0x
GPT-5.2 Pro	24.0x

The same prompt can therefore draw 80 times as much from your limit depending on the model, if you compare GPT-5.4 Mini with GPT-5.2 Pro. That is the actual learning effect of this indicator: you see, right at the point of choice, that a top-tier model with reasoning plays in a completely different league than a light model for routine work, and you make the decision consciously instead of by habit.

Briefly: what is the context window?

The context window is everything the model has in front of it at the moment of answering: your chat history so far, uploaded files, active Skills, and the instructions running in the background. Think of it as working memory for that single conversation. The space is limited, and everything you add takes up part of it. That fill level is exactly what the new indicator makes visible.

This is the biggest change. In the input bar you now see how much of your context window the current chat is occupying, for example as "13.6k of 200k" with a percentage. A click on it opens a breakdown showing where the space is going.

The main slots you can influence

• Messagesthe conversation so far. The longer the chat, the bigger this block.
• Attachments and imagesuploaded files. A large PDF can quickly take up most of the space here.
• Skills, memory, and custom instructionseverything Langdock automatically sends along with every message.
• Free spacewhat is still available before the window fills up.

You do not need to understand this in detail. It is enough to glance at it now and then. When free space gets tight or a single block stands out as unusually large, you know where to start: begin a new chat, remove a large file, or attach only the relevant section instead of the entire document.

You already know the third indicator in principle from last month's newsletter, and now it has become visible. You see your consumption in the current session window (five hours from the first message) and in the weekly window (reset every Monday).

This way, you notice early when you are moving faster than usual on a heavy morning, and you can consciously switch to a lighter model for the rest of the day.

Auto mode: convenient, with a small catch

Alongside the indicators, there is Auto mode, which takes model selection off your plate entirely. It analyzes your first message, estimates the complexity of the task, and picks a suitable model for the conversation based on that.

For anyone who does not want to think about the model at all, this is the simplest solution, and for most day-to-day work it makes good calls. Two points are still worth knowing. Because Auto uses the first message as its basis, you should jump straight into the actual task and not open with a brief greeting. And anyone who really wants to optimize consumption is often even more economical with conscious manual model selection, because Auto tends to reach for the stronger model when in doubt. Convenience and last-mile efficiency are a trade-off here.

The quiet cost drivers

Three things push consumption up without being immediately obvious, and those are exactly the ones the new context indicator finally makes visible:

Consumption used to be something you had to explain. Rules for switching models, reminders about large attachments, notes about the weekly limit, all things a team was supposed to learn and hold in mind. With the three indicators, that shifts.

What used to be knowledge is now delivered while you work. Whoever picks a model sees, right next to it, what that model costs. Whoever attaches a large file sees the context indicator jump. Whoever works intensely sees their session window grow. Abstract education turns into concrete feedback at the moment of the action.

Take a week where you consciously watch the indicators. Before sending, a quick glance at the model's consumption indicator; in a longer chat, an occasional look at the context indicator; and at the end of an intense day, a look at your session window.

After a few days, you will have a feel for which model fits which task and when a fresh chat beats the twentieth follow-up prompt.