Author: RadarAI Editorial
Editor: RadarAI Editorial
Last updated: 2026-11-18
Review status: Editorial review pending
Brief
速报
官方
AI动态
开源
At the Ignite 2026 conference, Microsoft unveiled its MAI series of models spanning speech, image, coding, and security, among which MAI Code 2 achieves about 96% of Claude Opus 5.5's performance at roughly 5% of the cost [15]; meanwhile, OpenAI launched the dual GPT-6.2 Astra/Sol models [16], and Google fully rolled out Gemini 4 Hydrogen across Search [19], intensifying the model race among the three major vendors in the enterprise AI market...
Editorial standards and source policy: Editorial standards, Team. Content links to primary sources; see Methodology.
## 🔍 Key Insights
At the Ignite 2026 conference, **Microsoft** unveiled its **MAI series of models** spanning speech, image, coding, and security, among which **MAI Code 2** achieves about 96% of **Claude Opus 5.5**'s performance at roughly 5% of the cost [15]; meanwhile, **OpenAI** launched the dual **GPT-6.2 Astra/Sol** models [16], and **Google** fully rolled out **Gemini 4 Hydrogen** across Search [19], further intensifying the model race among the three major vendors in the enterprise AI market.
## 🚀 Top Stories
- **Microsoft releases MAI Code 2 and MAI Code 2 Flash** [15]: Now available on GitHub Copilot, achieving about 96% of Claude Opus 5.5's performance at 5% of the cost.
- **Microsoft releases its first speech-to-speech model, MAI Realtime 1** [3]: Aimed at enterprise real-time voice scenarios, starting at $0.01 per audio minute.
- **Microsoft releases the MAI Image 2.7 series of image models** [10]: Topped the Artificial Analysis image editing leaderboard, at roughly $35/$10 per thousand images.
- **Microsoft releases the cybersecurity model MAI Cyber 2 Flash** [5]: 92.3% success rate on CyberGym-E2E, at about 2% of GPT-6 Astra's cost per task.
- **OpenAI releases GPT-6.2 Astra and GPT-6.2 Sol** [16]: The flagship model targets highly difficult projects, while the general-purpose model targets everyday work; both are now available on ChatGPT, Codex, and the API.
- **Gemini 4 Hydrogen fully rolled out across Google Search** [19]: Described as Search's largest model upgrade to date, strengthening intent understanding and multilingual capabilities.
- **Perplexity Computer adds a Multiagent tier** [13]: The main agent can dispatch up to 16 parallel agents to cross-check work.
- **After Anthropic's IPO, Amazon and Google together hold nearly 30%** [22]: Amazon holds 14.7% (potentially up to 23.8%) and Google holds 13.7%, all non-voting shares.
## 🔗 Sources
[3] Microsoft releases its first speech-to-speech model, MAI Realtime 1, starting at $0.01 per audio minute — https://aihot.news/items/xnvcva7n8z7451yy5802k9ewp
[5] Microsoft releases the cybersecurity model MAI Cyber 2 Flash, with a 92.3% success rate on CyberGym-E2E and a per-task cost of about 2% of GPT-6 Astra — https://aihot.news/items/ry5u3gdotnpbwhjtetzxchl49
[10] Microsoft releases the MAI Image 2.7 and 2.7 Flash image models, topping the Artificial Analysis image editing leaderboard — https://aihot.news/items/abyodbli5f89d0p6uc5lnuhsn
[13] Perplexity Computer adds a Multiagent tier: up to 16 AI agents cross-checking work — https://aihot.news/items/l8sxnqqa7lfnfj3rzdn8lj5l3
[15] Microsoft releases MAI Code 2 and MAI Code 2 Flash, achieving about 96% of Claude Opus 5.5's performance at 5% of the cost — https://aihot.news/items/vq6buq86aodbpij5oe6pe4arh
[16] Simon Willison tests OpenAI's new model GPT-
At the Ignite 2026 conference, Microsoft unveiled its MAI series of models spanning speech, image, coding, and security, among which MAI Code 2 achieves about 96% of Claude Opus 5.5's performance at roughly 5% of the cost [15]; meanwhile, OpenAI launched the dual GPT-6.2 Astra/Sol models [16], and Google fully rolled out Gemini 4 Hydrogen across Search [19], further intensifying the model race among the three major vendors in the enterprise AI market.
🚀 Top Stories
- Microsoft releases MAI Code 2 and MAI Code 2 Flash [15]: Now available on GitHub Copilot, achieving about 96% of Claude Opus 5.5's performance at 5% of the cost.
- Microsoft releases its first speech-to-speech model, MAI Realtime 1 [3]: Aimed at enterprise real-time voice scenarios, starting at $0.01 per audio minute.
- Microsoft releases the MAI Image 2.7 series of image models [10]: Topped the Artificial Analysis image editing leaderboard, at roughly $35/$10 per thousand images.
- Microsoft releases the cybersecurity model MAI Cyber 2 Flash [5]: 92.3% success rate on CyberGym-E2E, at about 2% of GPT-6 Astra's cost per task.
- OpenAI releases GPT-6.2 Astra and GPT-6.2 Sol [16]: The flagship model targets highly difficult projects, while the general-purpose model targets everyday work; both are now available on ChatGPT, Codex, and the API.
- Gemini 4 Hydrogen fully rolled out across Google Search [19]: Described as Search's largest model upgrade to date, strengthening intent understanding and multilingual capabilities.
- Perplexity Computer adds a Multiagent tier [13]: The main agent can dispatch up to 16 parallel agents to cross-check work.
- After Anthropic's IPO, Amazon and Google together hold nearly 30% [22]: Amazon holds 14.7% (potentially up to 23.8%) and Google holds 13.7%, all non-voting shares.
🔗 Sources
[3] Microsoft releases its first speech-to-speech model, MAI Realtime 1, starting at $0.01 per audio minute — https://aihot.news/items/xnvcva7n8z7451yy5802k9ewp
[5] Microsoft releases the cybersecurity model MAI Cyber 2 Flash, with a 92.3% success rate on CyberGym-E2E and a per-task cost of about 2% of GPT-6 Astra — https://aihot.news/items/ry5u3gdotnpbwhjtetzxchl49
[10] Microsoft releases the MAI Image 2.7 and 2.7 Flash image models, topping the Artificial Analysis image editing leaderboard — https://aihot.news/items/abyodbli5f89d0p6uc5lnuhsn
[13] Perplexity Computer adds a Multiagent tier: up to 16 AI agents cross-checking work — https://aihot.news/items/l8sxnqqa7lfnfj3rzdn8lj5l3
[15] Microsoft releases MAI Code 2 and MAI Code 2 Flash, achieving about 96% of Claude Opus 5.5's performance at 5% of the cost — https://aihot.news/items/vq6buq86aodbpij5oe6pe4arh
[16] Simon Willison tests OpenAI's new model GPT-
← Back to Updates