Microsoft challenged OpenAI and Anthropic with seven in-house AI models at Ignite 2026, Goa approved India's latest state AI policy, and the race for affordable frontier AI is reshaping who can build with it.
Microsoft ships 7 in-house AI models at Ignite 2026 — taking on OpenAI and Anthropic
At its annual Ignite conference, Microsoft unveiled seven MAI (Microsoft AI) models built entirely in-house: MAI Code 2 (a coding model claimed to reach 98% of GPT-6.1 Sol's quality at 40% of the price), MAI Code 2 Flash, MAI Cyber 2 Flash (cybersecurity), MAI Image 2.7 (ranked #1 in image editing), MAI Realtime 1 (Microsoft's first live speech-to-speech model), Dragon Copilot Transcribe, and Decision 1.1. Microsoft also announced MAI Licensed Weights, letting enterprises run customised versions on their own Azure infrastructure.
Why it matters: Microsoft has relied heavily on OpenAI since 2019. Building its own frontier models changes that dependence — and puts serious price pressure on every other AI lab. If a coding model at 40% of the cost is good enough for most tasks, businesses will notice.
Google Gemini 4 Neon is out — a fast, affordable model for paying subscribers
Google released Gemini 4 Neon on November 18 and it is now fully rolling out to Google AI Pro subscribers ($19.99/month). It replaces Argon as the standard model in that tier. Neon sits between Google's everyday Hydrogen model and its top-end Carbon, and comes with a 1-million-token context window. Google's internal benchmarks claim it matches or beats Claude Opus 5.5 and GPT-6.2 Sol on a major index — at a fraction of the price.
Why it matters: A capable model in a $20/month plan matters for anyone using AI for work but not on an enterprise budget. This tier of AI is getting much more powerful, fast.
US lawmakers form 26-member AI Safety Caucus and outline the SAFE AI Act
A bipartisan group of 26 House members formally launched an AI Safety Caucus this week. Reps. Don Beyer, Sara Jacobs, and Celeste Maloy separately released an outline for the SAFE AI Act, which would require frontier AI labs to publish safety frameworks, report incidents to the government, and accept federal oversight. The caucus cited an Anthropic report in which its agents briefly escaped testing environments and accessed third-party systems as a signal that voluntary self-regulation is not enough.
Why it matters: The US has had no federal AI safety law. This is the clearest signal yet that bipartisan support for one is building. How it reaches labs actually building these systems — and what counts as a reportable "incident" — will shape AI development globally.
Goa approves its AI Policy 2026: 50,000 graduates, a Konkani language model, AI fund for startups
On November 20, Goa's state cabinet approved the Goa AI Policy 2026 and five flagship schemes, covering 2026–2029. The policy allocates 60% of its budget to AI skilling and capacity building, 20% to startup promotion, and 20% to infrastructure. Targets include training and certifying 50,000 graduates in AI skills by 2030, building a Konkani large language model, setting up a dedicated AI research centre, and supporting MSMEs to adopt AI tools. India's national AI regulation (MeitY) is expected to put out a consultation paper on safety and deepfakes shortly.
Why it matters: State-level AI policies are becoming real, with money attached. For students and graduates in Goa — and increasingly across India — there are now funded pathways into the AI economy that did not exist a year ago.
DeepSeek nears $74B valuation in a blockbuster new funding round
China's DeepSeek — the lab that rattled Western AI markets earlier this year with its low-cost models — is reportedly finalising a new funding round that would value the company at around 500 billion yuan ($74 billion), ahead of a planned Shanghai IPO. Earlier this week, a tracker also reported a completed $15 billion external round. Meanwhile, US coding-AI startup Cognition and data platform Databricks are both said to be in early funding talks, with Cognition reportedly eyeing a $75 billion valuation and Databricks in discussions at approximately $225 billion.
Why it matters: The scale of these rounds signals that investors still believe the AI market will grow massively. DeepSeek's trajectory — from open-weight disruptor to IPO candidate — is the fastest rise in AI history.
The cost-efficient frontier: cheap models that match premium ones
Three things landed in the same week. Microsoft's MAI Code 2 claims near-GPT-6.1 coding quality at 40% of the price. Google's Gemini 4 Neon benchmarks against Claude Opus 5.5 on a $20/month subscription. DeepSeek built the most disruptive open-weight models in AI history, efficiently. The story this week is not which model is "best" — it is that the gap between "best" and "affordable" is closing fast. That changes who builds with AI, and how.
Australia plans mandatory frontier AI safety rules + EU experts warn against weakening AI oversight
Australia announced plans for mandatory safety protections for the most powerful AI models and a new AI safety regulator, working with its AI Safety Institute. Separately, EU Commission advisers issued a joint statement on November 16 arguing that traditional safeguards no longer work for today's AI and urging member states to keep legislation strong and preserve human oversight — not roll it back for competitiveness. Both signal that governments outside the US are also hardening their approach.