MAI-Code-1.1-Flash

Microsoft · Released Aug 2026

Conditional

MAI-Code-1.1-Flash is Microsoft's refreshed small-tier Copilot coding model — a quarter the cost and 22% better on Terminal-Bench 2.1 than its June predecessor, now with native vision. It's built for high-volume iterative work rather than frontier reasoning, making it the cheapest serious coding model in Copilot. Reach for it on everyday completions, refactors, and image-assisted debugging; use a flagship for complex architecture.

Is it right for you?

Good for

High-volume iterative Copilot workflows — completions, refactors, and scaffolds where latency and token cost dominate
Image-assisted debugging — native vision for reading screenshots, diagrams, and error images in chat
Cost-sensitive Copilot users — $0.20/$1.20 per 1M, 73% below the MAI-Code-1-Flash list price
.NET workloads — 15% improvement over MAI-Code-1-Flash

Not good for

Complex multi-file architecture or novel algorithm design — 138B total / 5B active sparse MoE caps reasoning depth
General-purpose non-coding tasks — a specialized coding model, not a general assistant
Independent benchmark verification — Microsoft published no task-by-task comparison, and vision attachment support is inconsistently documented across surfaces

How it performs by task

Terminal/CLI coding (Terminal-Bench 2.1)

Very Good

+22% over MAI-Code-1-Flash via GitHub Copilot CLI

.NET coding

Very Good

+15% over the predecessor

Code quality / survival

Very Good

+4% code survival through commit in production

Vision / image understanding

Good

Native support added, but rollout and attachment support are inconsistently documented across clients

Complex architecture / novel design

Fair

Small MoE (138B total / 5B active) caps frontier-level reasoning

Pricing

Input

$0.20 / 1M

Output

$1.20 / 1M

Context

256K context

View full pricing

Benchmarks

Benchmark	Score	Source
Terminal-Bench 2.1 (vs MAI-Code-1-Flash)	+22%	Source
.NET tasks (vs MAI-Code-1-Flash)	+15%	Source
Code survival through commit	+4%	Source

No verdict changes yet

The clock starts day one — changes land here as our verdict evolves.

Sources

Verification log

No verification checks yet

We haven't logged a verification check for this entry. Once a check runs, its history shows here.

How we evaluate