# Why Microsoft Thinks Smaller, Cheaper AI Can Beat Claude and GPT

- URL: https://talkfuse.com/story/why-microsoft-thinks-smaller-cheaper-ai-can-beat-claude-and-gpt/
- Published: 2026-11-17T23:16:03+00:00
- Updated: 2026-11-18T00:36:55+00:00
- Original publisher: Decrypt
- Original article: https://decrypt.co/383468/why-microsoft-thinks-smaller-cheaper-ai-beat-claude-gpt
- Topics: AI, Business, Companies, Cybersecurity, Technology

## Key points

- Microsoft is betting on specialised AI models that cost less to run.
- It claims strong coding and cybersecurity results against rival models.
- The models have limits, and some comparisons rely on company-run tests.

## Summary

Microsoft has launched specialised AI models for coding, cybersecurity and image generation, aiming to compete with OpenAI and Anthropic by offering lower costs for defined tasks. The strategy matters to businesses that run the same work repeatedly: a model focused on one job may be more economical while still performing well, though it is not necessarily better for general use.

Microsoft says MAI Code 2 reaches 96% of Claude Opus 5.5’s performance on four coding evaluations at 5% of its cost, and 98% of GPT-6.1 Sol’s performance at 40% of the price; the cost calculation uses a 50% promotional discount, and GPT-6.2 was released on the same day. MAI Cyber 2 Flash scored 92.3% on CyberGym-E2E vulnerability-discovery tasks, versus 91.5% for GPT-6 Astra, at under 2% of the cost per task, with 92 of 905 tasks excluded over training-data overlap. The coding models are already in GitHub Copilot, while the security model is due to begin replacing an older one for eligible MDASH customers in mid-December; Microsoft also says Code 2 is not yet suited to fundamental scientific research or very long autonomous assignments.

---
Summarised by TalkFuse (https://talkfuse.com/). Quote the original publisher for the reporting and link this page for the summary and discussion.
