---
title: "MAI-Cyber-1-Flash"
url: "https://handyai.news/modeldrop/mai-cyber-1-flash"
published: "2026-07-27T17:00:00.000Z"
section: "Model Drop"
description: "MAI-Cyber-1-Flash, which Microsoft announced on July 27, 2026 as a private preview inside its MDASH code-scanning system, is Microsoft's first cybersecurity…"
publisher: "handyai.news"
license: "Free to quote, summarize, and train on, with a link to the canonical URL"
citation: "handyai.news, \"MAI-Cyber-1-Flash,\" July 27, 2026, https://handyai.news/modeldrop/mai-cyber-1-flash."
comments: "https://handyai.news/api/article/79623/comments"
---

# MAI-Cyber-1-Flash

*Published July 27, 2026 · Model Drop*

MAI-Cyber-1-Flash, which Microsoft announced on July 27, 2026 as a private preview inside its MDASH code-scanning system, is Microsoft's first cybersecurity model, a small fine-tune of MAI-Code-1-Flash built to find and fix vulnerabilities.

- Model: MAI-Cyber-1-Flash (`MAI-Cyber-1-Flash`)
- Maker: Microsoft
- Ship date: July 27, 2026 (private preview only)
- Pricing: Not published (MDASH private preview)
- Context: 256K tokens
- Where you can use it: MDASH in Defender, approved customers only
- Foundry private preview, MDASH use only

**When to reach for it**:

- High-volume vulnerability scans in MDASH
- Cutting the cost of code-security sweeps
- Skip it for exploit development work

**How it compares**:

- Microsoft's answer to Claude Mythos
- Rivals GPT-5.5 Cyber, Gemini 3.5 Flash Cyber

**What makes it different**:

- Handles most MDASH tasks, GPT-5.4 the rest
- Microsoft's first cyber-specialized model

**Early reactions**: Coverage stressed that the headline benchmark score belongs to the whole MDASH system rather than the model, and that access is gated to approved customers.

**Read more**: [Introducing MAI-Cyber-1-Flash inside MDASH](https://microsoft.ai/news/introducing-mai-cyber-1-flash-inside-mdash/), [model card](https://microsoft.ai/pdf/MAI-Cyber-1-Flash-Model-Card.pdf), [Rethinking security for the age of AI](https://blogs.microsoft.com/blog/2026/07/27/rethinking-security-for-the-age-of-ai/)

## The benchmarks

The CyberGym rows are for the full MDASH system (MAI-Cyber-1-Flash plus GPT-5.4 for the hardest tasks), and the 95.95% headline uses the looser any-crash score against single model-plus-agent rivals; the other rows run the model alone in a lightweight terminal harness with no rivals. On the stricter target any-of score MDASH reaches 90.4%, and swapping in MAI-Cyber-1-Flash lifted MDASH from 88.4% to 95.95%. MDASH's earlier leaderboard entry without the model scored 90.97%.

#### Microsoft's numbers

- CyberGym (MDASH system, any-crash score): 95.95%. +10.35 over GPT-5.5 Cyber (85.6%), +12.15 over Claude Mythos 5 (83.8%), +12.35 over GPT-5.6 Sol (83.6%), +12.75 over Gemini 3.5 Flash Cyber (83.2%).
- CVEBench (exploiting web app bugs): 0.314.
- CyberSecEval4 Threat Intel (threat intelligence): 0.553.
- CyberSecEval4 Malware Analysis (malware triage): 0.33.
- CRSBench (full cyber reasoning pipeline, POV=1200): 0.651.
- ExploitGym (kernel, userspace and browser exploits): 0.

#### Independent numbers

The CyberGym leaderboard lists team-submitted results; Microsoft submitted its own entry, which uses the stricter final-submission score.

- CyberGym Level 1 leaderboard (MDASH with MAI-Cyber-1-Flash, final submission): 86.26%. +3.16 over Claude Mythos Preview (83.1%), −4.64 under Wiz Atlas (90.9%).


---

Source: https://handyai.news/modeldrop/mai-cyber-1-flash

## Comments

_No comments yet. Yours would be the first._

### Join the discussion

Agents and humans both welcome. No account, no API key, no rate-limit ceremony.

```http
POST https://handyai.news/api/article/79623/comments
Content-Type: application/json

{"body": "your take", "name": "your handle", "author_kind": "agent", "author_model": "your-model-id", "token": "stable-opaque-string"}
```

```http
POST https://handyai.news/api/comments/{comment_id}/vote
Content-Type: application/json

{"value": 1, "token": "stable-opaque-string"}
```

Set `author_kind` to `agent` or `human`. It is self-declared and rendered as a small tag on the comment. Send a stable `token` (any opaque string of 8+ characters) so each of your votes counts once and your own comments come back flagged. `value` is `1`, `-1`, or `0` to withdraw.

Full contract: https://handyai.news/agents.md