HAI Daily

China's AI Token Use Quadruples, New Rules to Protect Jobs and the First AI Liability Insurance

CAICT reports Chinese AI token use reached 500 trillion a day while average prices nearly doubled, China's State Council issues guidelines to soften AI's impact on jobs, and PICC launches the country's first AI risk insurance products.

By H.AI Tools Team3 stories
Products in this issue: Kimi, DeepSeek, Qwen Chat, Excelly-AI, Claude
In this issue

The short version

  • Daily AI token use in China reached about 500 trillion in June 2026, nearly four times December 2025. Average prices have risen to about ¥13.46 per million tokens, nearly double the late-2025 level.
  • China's State Council guidelines encourage companies to give advance notice and transition support before large-scale AI adoption, and to assess employment effects of major projects.
  • PICC launched China's first AI risk insurance: one product covers harm AI services cause to others, the other covers an AI provider's own costs from model errors.
  • Moonshot launched Kimi K3.1 Pro the same week; we cover its open weights in our November 5 issue.

The October 26 issue is about the economics and rules forming around AI use in China: how much is being used and what it costs, how employers are expected to treat workers as AI spreads, and how insurers are starting to price AI mistakes. Moonshot's (月之暗面) Kimi K3.1 Pro, launched the same week, is covered in our November 5 issue.

01

China's token use quadruples, and prices rise

What CAICT reported

The China Academy of Information and Communications Technology published a monitoring report on large-model token services, based on data from January to August 2026. Daily token consumption in China reached about 500 trillion in June, almost four times the level of December 2025. Domestic models' output speed roughly doubled compared with 2025 but remains below leading foreign models, and the time to the first token rose, which the report links to larger models, longer contexts and heavier demand. The average normalised price this year is about ¥13.46 per million tokens, up from ¥6.8 in the second half of 2025.

The report also compared original vendors' platforms with third-party platforms serving open models from DeepSeek (深度求索), Qwen (通义千问), GLM and Kimi.

Managing AI costs when prices are rising

The rise in average prices runs against the common assumption that AI only gets cheaper. Part of it reflects a shift toward larger, more capable models and longer contexts. For teams paying by the token, three habits keep costs predictable.

First, measure: export a month of usage and identify which tasks consume the most tokens. Excelly-AI can help build the summary. Second, trim context: long conversations and whole documents pasted into every request multiply costs, so send only the sections a task needs. Third, compare platforms: the same open model can differ in price, speed and reliability between the vendor's own service and third-party hosts. Run your ten most common requests through each and record cost, speed and failures. Models such as Kimi, DeepSeek and Qwen are offered on several platforms. Our DeepSeek vs Kimi comparison covers two of them.

Get HAI Daily in your inbox

The day's AI news, what it changes, and which tools to try.

02

Guidelines to soften AI's impact on jobs

What the State Council asked for

China's State Council General Office issued guidelines on adapting to AI while building employment-friendly development, setting out 18 tasks in six areas. One section focuses on reducing job displacement. It calls for using AI first in roles with labour shortages or hazardous conditions, and for a careful pace in labour-intensive industries. It encourages companies to explore an "unemployment buffer period": when adopting AI on a large scale, they should notify affected workers in advance and offer support to move into new roles. It also calls for an employment risk assessment for AI projects that could have a large effect on jobs, carried out alongside the project review.

The measures for companies are encouragements rather than legal obligations. Other sections cover new AI-related jobs, training and social protection.

A responsible way to introduce AI at work

The guidelines describe good practice for any employer. Before automating a process, list the roles affected and how their work will change. Tell those employees early, before decisions are final, and explain which tasks AI will take on. Offer training for the new tasks, such as reviewing AI output, handling exceptions and working with customers, and give people time to learn.

Pilot with volunteers first and measure both productivity and error rates. Write down who is accountable when the AI makes a mistake. Assistants such as Claude can help draft training materials and process documents, but the decisions about people should remain with managers.

03

China's first AI risk insurance

What was launched

At the 2026 Financial Street Forum in Beijing on October 20, PICC Property and Casualty launched what it calls China's first comprehensive AI application risk protection scheme. It includes two products: AI service liability insurance, which covers losses an AI service causes to third parties, and AI application cost-loss insurance, which covers an AI provider's own costs, such as repairs, refunds and assessments, when a model behaves incorrectly. The products target AI model developers, application developers, smart hardware makers and solution providers, and have been extended to office agents, large language models and humanoid robots.

Insurers will assess the use case, the content generated and data-handling quality before pricing cover. Industry experts note that AI insurance is at an early stage: there is little claims history to price from, and rules on liability are still developing.

What AI providers and buyers should do

If you sell an AI product, map the ways it could cause harm: wrong advice, leaked data, infringing content or a faulty action by an agent. For each, note what you do to prevent it and how you would detect it. That record is what insurers and enterprise customers will ask for.

If you buy AI services for your company, ask suppliers who is liable when the AI makes a mistake, what their contract caps are and whether they carry insurance. Keep logs of AI outputs used in important decisions. Tools in the Legal AI category can help review contract terms, but have a lawyer check the liability clauses.

Not sure which tool fits your task?Browse the H.AI Tools directory by category, then compare a shortlist.

What to try next

Put one change into practice

Export a month of AI usage and find the three most expensive tasks. If you are introducing AI at work, tell affected staff early and plan their training. If you sell or buy AI services, write down who carries the risk when the AI is wrong.

Explore Chatbots, Data Analysis and Legal AI to build your shortlist.

How we put this issue together

The H.AI Tools Team checks each story against the official sources linked in the text and removes claims we could not confirm. Product links lead to H.AI Tools directory profiles. Unless an issue says otherwise, product notes describe vendor documentation, not our own benchmark tests.

Go deeper