ProductNot yet confirmed elsewhere1 publisher2 min readPublished
Seven of eight Surface Laptop Ultra configurations lack the memory in Microsoft's local coding test
Microsoft's on-device coding model peaked at 75.5GB of memory at full context in the company's own test, according to a Command Line post. Only the 128GB model, at $5,899 and not yet on sale, has more, TweakTown reported.
The Product Desk
Key figures: the on-device MAI-Code-1.1 Flash peaked at 75.5GB of memory at full context in Microsoft's test; the quantized on-device model is 53GB; seven of eight Surface Laptop Ultra configurations have 64GB or less; the 128GB model costs $5,899; the base model costs $2,599.
| Figure | Value | Note | Claim |
|---|---|---|---|
| Peak memory at full context, Microsoft's test | 75.5 GB | 7 | |
| Quantized on-device model | 53 GB | 6 | |
| Memory, 7 of 8 configs | 64 GB or less | 13 | |
| 128GB model price | $5,899 | 9 | |
| Base model price | $2,599 | 14 |
What happened
- Nadella wrote in an X post on Microsoft's Windows and Surface event, "Your PC becomes an infinite software factory," and said Copilot can build software on the desktop with no cloud tokens.
- The on-device build of MAI-Code-1.1 Flash, a 137-billion-parameter coding model, is quantized to 53GB, about 80% smaller than the cloud version, per the Command Line post.
- The task-by-task routing between local and cloud ships at the end of October 2026, and Code is rolling out first through Microsoft's Frontier early-access program.
Why it matters
- contradiction The no-token pitch and the engineers' task-by-task routing disagree about where the work runs, so a team promising zero cloud use is promising more than Microsoft's docs do.
- cost Each seat that needs the 128GB model costs $3,300 more than the $2,599 base, so the memory requirement lands on the hardware budget.
- constraint Every performance figure is Microsoft's own, so a pilot has no outside benchmark to check the local-inference claim against.
- decision Whether to let non-developers use Copilot Code is a policy call that comes before any hardware order, and TechRadar says businesses have spent a year telling researchers they are worried about such tools.
Take an analyst on one of the seven Surface Laptop Ultra configurations with 64GB of memory or less [13], who asks Copilot Code for a dashboard. Microsoft's test of the on-device model peaked at 75.5GB at full context [7], so the analyst's machine has less than the test used [16]. The quantized weights alone are 53GB [6], and the peak sits 22.5GB above that [15].
TechRadar says Microsoft's own documentation shows the model runs best on a $5,899 laptop [12]. The memory figures support the narrower reading. The 128GB model is the only one of the eight with more memory than the test peak [16].
Nadella's pitch implies that a local model means a session with no cloud in it. Microsoft's engineers describe a router. Under GitHub Copilot's automatic mode, the choice between local and cloud is made separately for each task [3]. A caption on one diagram says "local inference does not make the session offline." [4] That routing ships at the end of October 2026, and Code is still limited to early access [5]. Any claim about how long people keep using what they build is a forecast.
We'd pilot on the 128GB seats and begin after the routing ships. The tradeoff is a small group on the most expensive hardware, plus a wait for that hardware to reach buyers. For the rest of the fleet, sort each seat on two facts: whether the machine has more memory than Microsoft's test peak, and whether the work carries a rule against leaving the device.
- More memory, cloud barred: the only cell where the pitch can hold. Test it once routing ships, because automatic mode decides per task [3]. - More memory, cloud allowed: the hardware premium buys local runs, and TechRadar expects harder tasks to route back to Azure [17], so cloud spend continues. - Less memory, cloud allowed: budget it as a cloud product with a local option. - Less memory, cloud barred: hold it. Microsoft's own caption says local inference does not make a session offline [4].
What to watch
- Whether the 128GB Surface Laptop Ultra goes on sale and independent testers reproduce Microsoft's 75.5GB peak and performance figures.
- Whether the end-of-October routing release lets admins keep chosen tasks on the device.
- Whether Copilot Code leaves Frontier early access and Microsoft publishes how long people keep using what they build.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence45
- Adoption
- Insufficient
- Hype gap+55
- Incentives60
- Confidence50
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
In a post on X summarizing Microsoft's recent Windows and Surface event, Nadella said Copilot can now build the software people need on the desktop without spending any cloud tokens.
- [2]
"Your PC becomes an infinite software factory."
- [3]
Microsoft's engineers say GitHub Copilot's automatic mode decides task by task whether work runs locally or in the cloud.
ReportedSupportedSource: Microsoft engineering documentation, as reported by TechRadarView cited source - [4]
One diagram caption in Microsoft's engineering documentation reads: "local inference does not make the session offline."
ReportedSupportedSource: Microsoft engineering documentation, as quoted by TechRadarView cited source - [5]
The automatic-mode routing does not ship until the end of October 2026; hybrid Copilot features reach Copilot+ PCs over the coming months, and Code is rolling out first through Microsoft's Frontier early-access program.
- [6]
The on-device version of MAI-Code-1.1 Flash, Microsoft's in-house 137-billion-parameter coding model, is quantized down to 53GB, roughly an 80% reduction from the cloud model.
- [7]
The on-device MAI-Code-1.1 Flash peaked at 75.5GB of memory at full context on Microsoft's test machine, a Surface Laptop Ultra built on Nvidia's RTX Spark.
- [8]
Eight Surface Laptop Ultra configurations are in play, and only one has more than 64GB of memory: the 128GB model.
- [9]
The 128GB Surface Laptop Ultra costs $5,899, which is $3,300 more than the base model, and is not yet available to purchase.
- [10]
All performance-related figures for the local model come from Microsoft.
- [11]
The businesses Microsoft most needs to convince have spent the past year telling researchers they are worried about citizen-built tools of this kind.
- [12]
Microsoft's own engineering documentation for the launch shows the factory runs best on a $5,899 laptop, still reaches into the cloud, and meets wariness from businesses.
- [13]
Seven of the eight Surface Laptop Ultra configurations have 64GB of memory or less.
- [14]
The base Surface Laptop Ultra costs $2,599.
- [15]
The 75.5GB test peak is 22.5GB above the 53GB quantized model.
- [16]
The 75.5GB test peak is above 64GB, so of the eight configurations only the 128GB model has more memory than the test peak.
- [17]
TechRadar argues that routing harder tasks to the Azure cloud conserves compute where possible while keeping Microsoft's cloud compute at the center.
Sources
1 independent publisher whose own reporting we read for this story.
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.
Topics
- AI Coding AssistantsFollow
- On-Device AIFollow
- Enterprise software adoptionFollow
- AI PCsFollow