Crypt0's NewsCrypt0's News

AI

Every Lab Shipped the Same Week and the Market Priced the Whole Field at Once

The frontier labs all chose the same week to ship, and the market answered by pricing the whole field in a single day. OpenAI released GPT-6.2 Astra and Sol on November 17, Microsoft unveiled seven new MAI models at Ignite the same day, Google made Gemini Tasks free across six countries, and Anthropic priced its public debut above its marketed range, with shares set to start trading Thursday. The models keep getting cheaper and sharper at the same time, and the contest has moved from who builds the smartest model to who puts it in the most hands at the lowest price. That shift is the story of the week.

OpenAI's pair covers both ends of the market in one release. Astra is the frontier model for the hardest professional and scientific work, with stronger results on multistep tasks using fewer output tokens, and OpenAI cut its price by 20 percent. Artificial Analysis scored Astra at 62 on its Intelligence Index, second only to Claude Fable 5.2 at 65, and BenchLeader ranked it third overall at 68.0. Sol is the efficient model for everyday coding and writing, holding its API token prices from GPT-6.1 Sol while adding speed, and Pro 200 subscribers get a new Ultrafast tier. Both models rolled out in ChatGPT Work, Codex, and the API on November 17, and OpenAI says many of its own employees already use Sol as their daily model.

Microsoft used its Ignite conference to answer with volume. Microsoft AI released seven models built to run agents around the clock at lower cost, including MAI Code 2 and MAI Code 2 Flash for coding, MAI Cyber 2 Flash for security work, MAI Image 2.7, and MAI Realtime 1. MAI Code 2 landed in GitHub Copilot the same day, with the Flash version in limited preview in Microsoft Foundry. The same conference made Word, Excel, and PowerPoint generally available inside the Copilot app, putting the full Office suite where the agents live. Microsoft is betting that the winning move is owning the workplace where the work happens, and it shipped seven models in a day to make the point.

Google answered with distribution instead of a new flagship. Tasks in Gemini went free for eligible users in the United States, Japan, Korea, Canada, Australia, and India, powered by Gemini 4 Hydrogen, the fast low cost model Google released on November 12. In the United States, Tasks can phone shops, pay with Google Pay, and take jobs by email or text, turning the assistant into an agent that finishes errands on its own. Google Labs also bought Howie, the scheduling assistant that lives inside email, for an undisclosed sum. The play is plain to see. Put a capable agent in front of hundreds of millions of people and let habit do the rest.

The market's verdict on all of it landed Wednesday evening. Anthropic priced its initial public offering at 83 dollars a share, above the 70 to 78 dollar range in its November 17 filing, according to CNBC and Bloomberg, with shares due to begin trading on the Nasdaq Global Select Market on Thursday, November 19, under the ticker ANT. The marketed terms cover 825 million Class A shares, plus a 26 billion dollar concurrent private placement backed by Google, Amazon, and Nvidia, and the reported price implies a valuation around 1.5 trillion dollars on a fully diluted basis. Anthropic reported third quarter revenue of 17.0 billion dollars, more than 12 times the year earlier level, and the filing still shows the 70 to 78 dollar range, so the 83 dollar figure stands as reported terms until the final prospectus lands. The first frontier lab to list publicly did so at a price the market pushed higher.

The tooling around the models moved just as fast. Anthropic added an experimental Code Mode to Claude Code that chains tools into scripts and keeps answering while long commands run, and opened a faster release channel for Claude Desktop. Linked billing, now in beta, lets tools such as Linear, Ramp, and Factory bill a Claude organization directly, so teams pay for the work instead of passing keys around. The open weight bench added Rumble 2 Flash on November 18, keeping the open frontier in step with the closed one. Every layer of the stack, from the model to the billing, is being rebuilt for agents that run on their own for hours at a time.

For readers choosing tools this week, the lesson is simple and practical. Pick the model by the job and the price, because the frontier now ships in pairs, a flagship for the hardest work and a cheaper sibling for everything else, and the real difference sits in distribution, Tasks free in six countries, Office inside Copilot, agents that pay and schedule on their own. Anthropic's debut puts a public price on the whole field, and that price says the market believes the agent era is still early. Try the free tiers first, measure what each model does for your actual work, and let the results pick your stack.

Quick answers

What is this story about?

The frontier labs all chose the same week to ship, and the market answered by pricing the whole field in a single day. OpenAI released GPT-6.2 Astra and Sol on November 17, Microsoft unveiled seven new MAI models at Ignite the same day, Google made Gemini Tasks free across six countries, and Anthropic priced its public debut above its marketed range, with shares set to start trading Thursday. The models keep getting cheaper and sharper at the same time, and the contest has moved from who builds the smartest model to who puts it in the most hands at the lowest price. That shift is the story of the week.

Why does this story matter?

For readers choosing tools this week, the lesson is simple and practical. Pick the model by the job and the price, because the frontier now ships in pairs, a flagship for the hardest work and a cheaper sibling for everything else, and the real difference sits in distribution, Tasks free in six countries, Office inside Copilot, agents that pay and schedule on their own. Anthropic's debut puts a public price on the whole field, and that price says the market believes the agent era is still early. Try the free tiers first, measure what each model does for your actual work, and let the results pick your stack.

Sources

New to crypto? Read the crypto glossary, browse frequent questions, read our story, or explore the story archive.

← Back to Crypt0's News