AI
Small Models Got Cheap, ChatGPT Grew a Workshop, and the Agent Race Moved Into the Office
October 8 was the day the model became the cheap part and the agent became the product. Anthropic slashed its small model pricing by up to 90 percent, OpenAI turned ChatGPT into a workshop that builds interactive tools inside the chat, and Google launched a work agent that lives inside the inbox. Three separate launches, one shared thesis. Intelligence keeps getting cheaper by the week, and the competition has moved to who puts it to work.
Anthropic fired the pricing shot first. Claude Haiku 5.5, released October 7, costs $0.10 per million input tokens and $0.50 per million output for prompts under 100,000 tokens, matching OpenAI's GPT 6 Luna at exactly that tier and running roughly 90 percent cheaper than Haiku 4.5 on the most common usage band. The model brings an adjustable effort dial, Low through Max, to the Haiku line for the first time, alongside a 1 million token context window and stronger scores on computer use and agent benchmarks. Anthropic also halved Sonnet 5.5 cache read prices and started issuing $100 to $500 in monthly API credits to Max and Team subscribers. The timing speaks volumes. Anthropic hosts IPO investors on October 14, and the cheapest credible small model in the market is a strong calling card.
OpenAI answered with distribution. GPT 6 became the default model across ChatGPT on October 7, reaching the free tier on October 8, which puts the upgrade in front of a large share of the service's 1.2 billion weekly users. The headline feature is Intelligent UI, which renders answers as checklists, cards, charts, forms, and small working tools instead of plain text. GPT 6 Instant begins responding 44 percent sooner on average than GPT 5.6 Instant for questions requiring web searches. The same week, OpenAI published 722 mathematics manuscripts, including a proof of the Unique Games Conjecture, a result on Barnette's Conjecture, and progress on four of the seven Millennium Prize problems, with mathematicians including Terence Tao weighing in on the results. The message is scale in every direction, cheaper to run, faster to answer, and ambitious enough to attack the hardest problems in math.
Google made the office its battleground. Google Cloud introduced the Gemini agent on October 8, a single AI agent that plans work, uses tools, connects to company systems, and returns finished work in documents, email, and developer environments. It works inside Google Workspace apps like Gmail, Docs, Sheets, and Calendar, and also through Microsoft 365 and Slack. It picks the best model for each task from Google's Gemini models and Anthropic's Claude models, with more to come. Companies can create coworker agents that act as team members with their own email address and access only to the information they are given. Versions for financial services and legal work are already in preview, with government, healthcare, and retail versions on the way. The launch follows OpenAI's September rollout of always on agents and Meta's Muse personal agent, so the workplace agent race now has three serious runners.
The independents are raising like it is still 2021. Manus, the AI agent startup, raised more than $500 million after its proposed acquisition by Meta unraveled, according to The Information's October 8 report. The company builds agents that carry out multistep tasks across applications, from research to document preparation, and the round gives it the resources to stay independent while the giants integrate similar capabilities into their platforms. Behind the software, the industrial buildout kept pace. Microsoft shipped Nvidia powered Surface AI PCs that run models above 120 billion parameters locally, starting at $2,600, alongside Microsoft Execution Containers, a Windows 11 sandbox that defines exactly which files and networks an agent can touch. TSMC's September revenue jumped 54.6 percent year over year, and GlobalFoundries signed a five year deal to build chip packaging components for TSMC in New York. The agents are coming, and the factories are already pouring the concrete.
Put the week together and the shape of the next phase is visible. The frontier models keep getting more capable, the small models keep getting radically cheaper, and the value is moving to the layer that does the work. For the reader, the practical read is simple. Learn to orchestrate agents and small models now, while the pricing is generous and the tooling is fresh, because the companies selling intelligence have decided the future is labor, and they are pricing accordingly.
Quick answers
What is this story about?
October 8 was the day the model became the cheap part and the agent became the product. Anthropic slashed its small model pricing by up to 90 percent, OpenAI turned ChatGPT into a workshop that builds interactive tools inside the chat, and Google launched a work agent that lives inside the inbox. Three separate launches, one shared thesis. Intelligence keeps getting cheaper by the week, and the competition has moved to who puts it to work.
Why does this story matter?
Put the week together and the shape of the next phase is visible. The frontier models keep getting more capable, the small models keep getting radically cheaper, and the value is moving to the layer that does the work. For the reader, the practical read is simple. Learn to orchestrate agents and small models now, while the pricing is generous and the tooling is fresh, because the companies selling intelligence have decided the future is labor, and they are pricing accordingly.
Sources
New to crypto? Read the crypto glossary, browse frequent questions, read our story, or explore the story archive.