Models & Products 22 September 2026
Anthropic releases Claude Opus 5.5 at a lower price
Anthropic released Claude Opus 5.5, its new top model. The company says it is better than Opus 5 at coding, operating a computer and office work, and writes more than 30% faster. The price fell 20%, to $4 per million input tokens and $20 per million output tokens. It is available on Anthropic's own platform and through AWS, Google Cloud and Microsoft Azure.
Why it matters: Indian firms that build products on Claude get a stronger model for less money. Anthropic says typical workloads cost about 40% less than on Opus 5.
Models & Products 22 September 2026
OpenAI launches GPT-6 Sol and GPT-6 Luna and halves prices
OpenAI released GPT-6 Sol, for harder work such as coding, and GPT-6 Luna, for simple high-volume tasks such as summarising documents. OpenAI says the GPT-6 models cost half as much as the GPT-5.6 models. It also says Sol makes about half as many mistakes as the model it replaces. Luna is rolling out to Free and Go users of ChatGPT.
Why it matters: Free and low-cost ChatGPT Go users in India get access to a newer model. Businesses that pay for OpenAI's API can run routine tasks such as extracting data from documents more cheaply.
India 24 September 2026
Government considers ₹15,000 to 20,000 crore fund for frontier AI
The government is considering a National Frontier AI & Compute Fund, with a government contribution of ₹15,000 to 20,000 crore. The fund would invest in companies building large AI models, in GPU clusters and in data centres. It would likely be set up as a SEBI-regulated Category I Alternative Investment Fund that also brings in private money. The idea is still under consultation, and the final size and structure have not been decided.
Why it matters: If approved, the fund would give Indian AI companies long-term money that ordinary grants and venture funding struggle to provide. That could mean more AI tools built in India for Indian languages and needs.
India 25 September 2026
Sarvam AI releases Vision 2.1 for documents in 22 Indian languages
Bengaluru-based Sarvam AI released Vision 2.1, a model that reads scanned documents in 22 Indian languages. It handles tables, form fields and handwritten text. It is offered through two services: one turns documents into digital text, and the other pulls out specific data. Sarvam reported a score of 87.30 on the olmOCR-Bench test, ahead of the other models it compared against.
Why it matters: Much Indian paperwork is in regional languages or handwritten. Offices, banks and small businesses could use tools like this to digitise forms and records without typing them in by hand.
India 24 September 2026
Microsoft opens Hyderabad cloud region; Telangana targets 5 GW of data centres
Microsoft opened its India South Central cloud region in Hyderabad, its fourth cloud region in India. Microsoft says the facility is built for AI workloads and uses almost no water for cooling. The Telangana government said it aims for 5 GW of data centre capacity by 2029.
Why it matters: More data centres in India let businesses and government bodies keep data in the country and run AI services closer to their users. The investment also brings construction and technical jobs.
India 24 September 2026
Odisha approves ₹1.45 lakh crore of data centre and AI projects
Odisha's High-Level Clearance Authority, chaired by Chief Minister Mohan Charan Majhi, approved technology projects worth ₹1.45 lakh crore. Most of the money is for data centres in Khordha, Puri and Cuttack, including an AI data centre by HCL Technologies. A semiconductor unit was also approved. The state expects more than 5,100 jobs from the projects.
Why it matters: It shows AI infrastructure spreading beyond the usual tech cities to states such as Odisha, which could bring jobs and local business opportunities there.
India 22 September 2026
SEBI uses AI to catch market manipulation and fake stock tips
Reports described how SEBI uses its in-house AI system, Sudarshan, to scan trading data for unusual patterns. It also matches stock tips posted on X, Instagram and Telegram against actual trading, to find pump-and-dump schemes. SEBI sends about 7,000 takedown requests a month, and more than 100,000 videos have been removed. SEBI officials still check AI findings before taking any action.
Why it matters: Retail investors who follow stock tips on social media should know that SEBI now watches these channels closely. It also shows how an Indian regulator uses AI while keeping humans in charge of decisions.
Models & Products 25 September 2026
Microsoft redesigns Copilot with an agent that keeps working on its own
Microsoft announced a new version of its Copilot work assistant with three parts. Home puts chat and Word, Excel and PowerPoint in one place. Code lets people describe an app or automation in plain words and have Copilot build it. Autopilot is an agent that keeps working on a set goal, such as supplier reviews, without being prompted each time. Microsoft is also moving to two kinds of billing: fixed subscriptions for everyday use, and pay-per-use for heavier agent work.
Why it matters: Many Indian offices use Microsoft 365, so these tools may reach their staff first through the Frontier early-access programme. Firms should watch the new pay-per-use billing to keep costs under control.
Models & Products 24 September 2026
Meta shows Muse Charm, a keychain device for its AI agent
At its Connect 2026 event, Meta showed Muse Charm, a keychain-sized device with a small screen, a fingerprint sensor and voice control. It gives users access to Meta's Muse AI agent, which can book travel, manage email and make purchases. Meta plans to release it in December 2026 but has not given a price or said which countries will get it.
Why it matters: Big companies are building AI assistants that act for users, not only answer questions. Muse is already offered on WhatsApp in some countries, which matters in India, where WhatsApp is very widely used.
Policy & Rules 21 September 2026
US proposes AI incident alert system to China
US Treasury Secretary Scott Bessent said the US proposed a notification system with China for AI incidents that affect national security. He raised it in talks with Chinese Vice Premier He Lifeng in New York, ahead of a meeting between President Trump and President Xi. It is a proposal, not a signed agreement.
Why it matters: The two largest AI powers talking about AI safety could shape the global rules that India and other countries later follow.
Safety & Security 22 September 2026
British Columbia sues OpenAI over Tumbler Ridge school shooting
The Canadian province of British Columbia sued OpenAI and its CEO Sam Altman in a federal court in San Francisco. It says OpenAI did not alert police about threats the shooter made on ChatGPT before the February 2026 attack, in which eight people were killed. The province wants compensation and a court order forcing OpenAI to change how it handles conversations that threaten violence.
Why it matters: Governments are starting to hold AI companies legally responsible for what happens on their chatbots. The outcome could shape the safety rules these companies follow in every country, including India.
Compiled from public reports and official announcements. Follow the source links for full details. Next update: Saturday, 17 October 2026, 2:00 PM IST.