- BIST10012,151-0.94%
- USD49.220.00%
- EUR55.21+0.28%
- GOLD4,195.90+0.13%
- INTEREST37.00%-1.00
- BITCOIN82,230-1.72%
- October 13, 202618°
- BIST10012,151-0.94%
- USD49.220.00%
- EUR55.21+0.28%
- GOLD4,195.90+0.13%
- INTEREST37.00%-1.00
- BITCOIN82,230-1.72%
- October 13, 202618°
Microsoft introduced its own artificial intelligence models
Microsoft announced the Majorana 2 quantum chip along with seven new artificial intelligence models at the Build conference. The company aims to reduce dependency on OpenAI and Anthropic...
Microsoft introduced a new family consisting of seven AI models developed in-house at its annual Build developer conference in San Francisco. This move stood out as one of the clearest steps toward reducing the company's reliance on external AI companies, into which it has invested billions of dollars.
Mustafa Suleyman, head of the company's AI division, stated that the models surpass OpenAI's GPT-5.5 model in terms of quality. Suleyman also noted that based on publicly available pricing data and calculations according to different model sizes, they anticipate about a tenfold cost advantage.
Highlights
- Microsoft introduced seven new AI models at the Build conference.
- The company aims to reduce dependency on external model providers.
- MAI-Thinking-1 was announced as Microsoft's first reasoning model.
- MAI-Code-1-Flash focuses on converting written descriptions into code.
- The Majorana 2 quantum chip was presented as more reliable than the previous generation.
Focus of the new models
The standout model in Microsoft's new family was MAI-Thinking-1. The company announced that it trained this model from scratch using clean and commercially licensed data. It was emphasized that the model was developed without distillation from third-party systems.
MAI-Thinking-1, which has 35 billion active parameters, operates with a wide context window of 256,000 tokens. The model is designed for multi-step complex tasks, long-text reasoning, and code generation.
The conference also introduced the MAI-Code-1-Flash model. This model focuses on converting text descriptions into source code for applications and websites. Microsoft reported that this model is available for use across GitHub Copilot and Visual Studio Code.
Message of competition to OpenAI and Anthropic
Microsoft plans to run its own models on the Azure infrastructure to reduce fees paid to third-party providers and pass these savings on to developers. The company stated that in blind tests conducted by independent human evaluator partner Surge, MAI-Thinking-1 was preferred over Anthropic's Claude Sonnet 4.6 model.
The company also claimed that in coding benchmarks, the model performed on par with Claude Opus 4.6. These statements reveal Microsoft's desire to be not only a user but also a direct model producer in the AI field.
New milestone in quantum chips
At the Build event, not only AI models but also new developments in quantum computing were announced. Microsoft stated that the Majorana 2 quantum chip has become a thousand times more reliable than its predecessor.
The company noted that this is an important threshold on the path to a commercially viable quantum computer. It was explained that qubits used in quantum computing are extremely sensitive to temperature and vibration, and the new chip directly targets this issue.
While the qubits in the new chip were limited to milliseconds in the first generation, they can now remain stable for an average of about 20 seconds. Microsoft compared this progress to moving from a device that requires daily charging to one that can last for years.
Zulfi Alam, corporate vice president of Microsoft Quantum, said they expect to reach a quantum machine capable of solving commercially meaningful and reasonable problems by 2029. It was noted that the chip currently contains 12 qubits, while a truly useful system will require millions of qubits.
Scientific debates continue
Microsoft's quantum approach is based on so-called topological qubits theorized by Italian physicist Ettore Majorana in the 1930s. However, the company's work in this area has previously sparked debate; a 2018 study published in Nature claiming to provide evidence for this particle was retracted.
In the second-generation chip, lead was used instead of aluminum as a partially superconducting material, which was said to improve upon the first version. However, the chip and supporting research have not yet undergone peer review, and some physicists argue that more data should be shared.
Coincided with market dynamics
Microsoft's shift to its own models coincided with the AI companies it has committed billions of dollars to preparing for public offerings. Anthropic filed a confidential initial public offering on June 1 after its Series H round raised $6.5 billion, valuing the company at $96.5 billion.
OpenAI is also reportedly preparing a similar confidential filing. Microsoft has committed up to $13 billion to OpenAI and up to $5 billion to Anthropic; both companies' models are offered to customers via Azure.
Context: Microsoft's new model family shows the company's desire to move beyond being just a customer network of large external partners in the AI race and to compete directly with its own systems. At the same time, the announcements in the quantum field indicate that the company is also strengthening its position in the long-term technology race.
Stay Updated with The Turkiye Times
Subscribe to be informed about special news, developments and announcements. We will keep you informed.
By signing up, you agree to our Privacy Policy
