The model you pick in Copilot Cowork can nearly triple what a task costs. I ran the same task on three of the models in the picker: Claude Sonnet 4.6 cost $3.98, Claude Opus 4.8 cost $4.77, and GPT 5.5 cost $10.69. The output was the same quality across all three. When I re-ran the comparison in July on newer models, the cheapest model changed, and I’ve added those results further down.
Microsoft made Copilot Cowork generally available on 16 June 2026, and every Cowork run now spends Copilot Credits. The labels in the model picker tell you what each model is good at. They don’t tell you what each one costs, and that’s the part that lands on your bill.
How the Copilot Cowork model picker works
When you start a Cowork task, the picker gives you four options. Each produces broadly similar output, but the pricing is very different.
General availability moved Cowork out of the Frontier preview and onto consumption billing. Every task spends Copilot Credits, charged at $0.01 per credit on pay-as-you-go. The model is one of four things that drive the cost of a task, alongside context retrieval, tool calls and runtime, and it’s the only one you choose directly.
The four options are:
- Auto picks the model it thinks suits your task. It’s the default.
- Claude Sonnet 4.6 is labelled “efficient for everyday tasks”.
- Claude Opus 4.8 is labelled for “complex, high-stakes work”.
- GPT 5.5 is labelled “versatile across task types”.
Microsoft’s guidance in Choose a model for Copilot Cowork is to leave the picker on Auto for most day-to-day work. That’s reasonable advice, but it’s also where most people stop thinking about it.
The test: one task, three models
I ran the identical Cowork task three times on three models and asked for the same outputs each time. The cheapest run cost $3.98 and the dearest cost $10.69.
The task started from a CSV of 2025 sales figures and asked for three things: a multi-tab Excel workbook, a board-ready PowerPoint deck and an interactive HTML dashboard. After each run I used the /cost command to read the credits consumed.
| Model | Credits | Cost (pay-as-you-go) |
|---|---|---|
| Claude Sonnet 4.6 | 398 | $3.98 |
| Claude Opus 4.8 | 477 | $4.77 |
| GPT 5.5 | 1,069 | $10.69 |
GPT 5.5 cost 2.7 times as much as Sonnet for equivalent work. Opus sat surprisingly close to Sonnet, about 80 cents more for the deeper reasoning. The model labelled “versatile across task types” was the outlier by a long way.
The output quality was equivalent across all three runs. I used a custom skill called interactive-dashboard that holds the output design constant, so the difference in credits comes from the model rather than from different work being done.
What the picker labels don’t tell you
Read the labels again: “efficient”, “complex, high-stakes” and “versatile”. If you were choosing the safe middle option from those three, you’d probably pick “versatile”.
In my test, GPT 5.5 cost about the same as running Sonnet and Opus combined. Microsoft’s Copilot Credits Guide doesn’t break consumption down by model, so the label ends up doing most of the steering.
That’s why I think of the picker as a behaviour design decision. If your people treat “versatile” as their default, you’ll spend two to three times what you need to on work Sonnet handles just as well. Auto is a sensible default if you don’t want people choosing at all. Sonnet is the better default if you want predictable everyday cost.
This is the same pricing trap that catches teams with licensing, one layer deeper. I covered the licensing side in Microsoft 365 Copilot vs Copilot Chat.
Which Copilot Cowork model to use for which task
Use Sonnet for routine work. Drafting, summarising, restructuring data and producing standard documents from a brief. This is the “everyday tasks” work Microsoft describes, and it’s where most of your Cowork runs will sit.
Save Opus for work where a reasoning slip is expensive. Multi-source analysis, nuanced research synthesis, or anything where a mistake costs more than the premium. On most jobs the difference between Sonnet and Opus is under a dollar. Sometimes the output is noticeably better, often it isn’t, so test it on your own work before making it the default.
Use Auto when you don’t want to choose. Auto picks a model per task and tends to lean towards the cheaper end. Microsoft’s general availability announcement positions it as the right default for mixed workloads, and it works well paired with a custom skill that fixes the shape of the output.
Treat GPT 5.5 as a specialist. It has a place for long-form writing, citation-heavy outputs and very large context, but it shouldn’t be anyone’s default.
Update: what the July tests showed
The model list kept moving after this test, so I kept testing. In July I ran two more comparisons on a heavier job: ticketing data in a SharePoint site, with an interactive HTML dashboard, a PowerPoint deck, a Word summary and a drafted email from one prompt.
Claude Sonnet 5 against Claude Fable 5 came in at 912.2 credits and 1,738.4 credits. Fable 5 added a sticky header on the dashboard and a bit more colour, which were real differences, but I couldn’t see 826 credits’ worth of them.
Then GPT-5.6 arrived in Cowork, so I ran the same prompt across all six models available at the time:
| Model | Credits |
|---|---|
| Claude Opus 4.8 | 798 |
| Claude Sonnet 5 | 1,175 |
| GPT-5.6 Terra | 1,781 |
| GPT-5.5 | 1,992 |
| Claude Fable 5 | 2,128.5 |
| GPT-5.6 Sol | 9,324.7 |
The most expensive run cost 11.7 times the cheapest for comparable output. Opus 4.8 produced the best dashboard and was also the cheapest on this job, which is the opposite of what the June test would lead you to expect. Terra beat GPT-5.5 on both cost and quality, so if you’re routing work through GPT-5.5 today, Terra is the better choice.
The lesson I take from all three tests is the same. The cheapest model changes with the task and with each new release, so don’t set a default once and forget it. Test your common tasks on two or three models whenever the list changes.
How to keep Cowork costs under control
Copilot Credits are a new kind of budget line, and the model is the lever that decides what the meter reads. Run the same prompt on two models, check /cost after each, and you’ll quickly build a feel for what your typical tasks cost. One click in the picker doesn’t look like much, but across a team and a month it adds up.
Frequently asked questions
What models are available in Copilot Cowork? The picker offers four options: Auto (the default, which chooses for you), Claude Sonnet 4.6 for everyday tasks, Claude Opus 4.8 for complex, high-stakes work, and GPT 5.5 for versatility across task types. Sonnet with Opus as an advisor is also available as a paired mode in some tenants. What you see depends on what your organisation allows.
How much does each Copilot Cowork model cost? There’s no fixed price per task. Cowork bills in Copilot Credits at one cent per credit on pay-as-you-go. In my test of the same task, Sonnet 4.6 used 398 credits ($3.98), Opus 4.8 used 477 ($4.77) and GPT 5.5 used 1,069 ($10.69). Prompt complexity, context retrieval, tool calls and runtime all change the total.
Which Copilot Cowork model should I use by default? Auto if you’d rather not think about it. Otherwise, test your most common tasks on two or three models and pick on evidence, because the cheapest model changes with the task. In June, Sonnet 4.6 was cheapest on my test. In July, on a heavier job, Opus 4.8 was cheapest and produced the best output.
How do I check what a Cowork task costs?
Type /cost in the task window after a run finishes. Cowork shows the number of Copilot Credits the conversation has used so far. Run it after each model trial to compare like for like.
