token

ON OCTOBER 26, A JOINT PUBLICATION WAS MADE BY THE CHINESE INSTITUTE OF ARTIFICIAL INTELLIGENCE, AMONG OTHERS. THE REPORT SHOWS THAT, AFTER TWO YEARS OF PRICES, DOMESTIC AND FOREIGN PRICES FOR LARGE MODEL SERVICES ROSE SIGNIFICANTLY THIS YEAR. OF THESE, THE DOMESTIC AGGREGATE AVERAGE PRICE OF LARGE MODEL SERVICES INCREASED FROM $6.8 IN THE SECOND HALF OF 2025 TO $13.46 SINCE THE PRESENT TIME, CLOSE TO $981 TP3T. IN CONTRAST, THE FOREIGN MODEL PUBLISHED SINCE 2026 HAS AVERAGED A UNIFORM PRICE OF ABOUT $117.8, STILL CLOSE TO HOME..

Message dated 22 September, yesterday, SpaceXAI officially released Grok 4.7, the main programmer, long mission and professional expertise. The new model has now entered Cursor, Grok Build and Grok API and will be provided through a third-party programming tool, model route platform and cloud service. According to the official presentation, Grok 4.7 uses a new, larger base model than Grok 4.6 and extends the length of intensive learning and training, focusing on long context management, self-censorship and the ability to perform complex tasks for hours..

On September 9, according to the news, DeepSeek will re-pricing the flash series at 12 noon on September 10, with the input price falling from $1.5 ton to $1 per million token, the output price falling from $4.5 to $4 per person per person per person per person per person per person per person per person per person per person per person per hour, and the cache price from 0.05 to 0.02 dollars per person per person per person per person per person per month. At the same time, the middle version of DeepSeek V4.1 Flash has been open-ended. API users participating in internal measurements need not change b..

On September 4th, Skycat went online on September 3rd, the AI Space Station, the Token Filling Centre. The first of the pages, which brought together Aliyun, Gens, Kimi and MiniMax to subscribe to the products, can be purchased directly by Skycats. Goods in sale include periodic subscriptions to Token Plan, Coding Plan, and a charge-to-use scheme to support exchange codes and direct delivery. The day before this, Ji-Shu had opened the official flagship shop and built a subscription package like Coding Plan..

On September 3rd, according to the news, on September 2nd, an official flagship shop was opened in Skycat with a standard seat for the GMCing Plan personal Lite, Pro, Max and Team. The main subject of the shop is Beijing Psychiatry and Technology Co. Ltd., which supports the purchase of meals on a monthly, quarterly or annual basis. The package is based on GM-5.3 and can be used for more than 20 Agent programming tools such as ZCode, Claude Code, Codex and others. Personal Lite pays $118 a month..

On August 14th, according to The Insider of Business, the Canadian American investor Y Combinator CEO Garry Tan suggested in a podcast released yesterday that the founders of the start-up company should not overwhelm AI Token's consumption costs, but should adopt the "Tokenmaxing" strategy to gain a competitive advantage by investing heavily in AI intelligence. Garry Tan says that the founders should allow their own use of AI Token in large quantities, even if this entails high expenditures. He thinks through..

On 1 August, according to Krypton, Cognition Engineer and former Meta Engineer Marpe Won stated in an interview that Silicon Valley AI entrepreneurship had reached the stage of rapid consensus failure. The reduction in middle-level management and the increase in the number of general-level engineers are not just organizational flattening slogans, but part of the management was redivided after AI began to assume responsibility for information collation, synchronization of progress and cross-group collaboration. He is particularly opposed to using Token consumption to measure whether an engineer is "AI native". Open rankings induce employees to run empty for numbers, leaving only higher..

On July 22, Google released Gemini 3.6 Flash, 3.5 Flash-Lite and Code-oriented Security 3.5 Flash Cyber today. Gemini 3.6 Flash reduces 171 TP3T output Token by 3.5 Flash and reduces the number of reasoning steps and tool calls in multistep workflows. API pricing is token 1.5, output token 7.5, DeepSWE..

On July 14, according to news, the free version of Hy3 has come to the top of OpenRouter LLM week. On July 6th, the telecommunication mixer officially launched the Hy3 model and subsequently launched a two-week free API call on the AI Model Convergence platform OpenRouter. According to the data, the free version of Hy3 processed 6.13 trillion token per week, the highest on the weekly list. This figure exceeds the 5.95 trillion Token of MiMo-V2.5..

    Read guidance: Below is a directory where readers can jump-start and read. If you are interested in Token ' s calculations and charges, then 4-7 is worth reading. Target for writing: Handwritten over 7,000 words, long space, suggestions for reading, which can be saved first. Why write by hand, and try not to look at other books, because I want to simplify the information through my own thinking to those who want to know about AI, not just the professionals, but to refer to other books, it is inevitable that professional terms are inadvertently revealed. 1. Purpose of writing (...)

Token is what it is, and Token has become a new base of economic elements and strategic resources. It's written at the beginning of the wave of the AI wave, with applications coming out of place, and it's changing. No one will doubt the correctness of the phrase “AI will go into the lives of all of us”. But in all the discussions about AI, you must have heard a word very often — “Token”. So, what the hell is Token? Token is a very central concept in computer and Internet technology. Token, we need to know that Token's three different things..

On 3 July, according to CNBC, Palantir CEO Alex Karp, during an interview, strongly criticized the Token billing model used by the AI Laboratory, including OpenAI and Anthropic, and stated that the structure was facing serious problems. Karp points out that as AI’s use costs soar, business is increasingly frustrated with this “token maxxing” business model. Karp believes that as the new model is more expensive than the previous version

According to information received on 2 July, HKTIP documents that Beijing Silicon-based Mobile Science and Technology Ltd. has officially submitted its listing application to Hong Kong and that the co-prospects were WATCH International and the State of Tai Haitong. As at 30 April this year, the number of subscribers to Silicon-based Mobile Platforms had reached 102,824; the average daily consumption of Token in April this year was 57,855 million. In 2025 Token's annual throughput, Silicon-based flows are China's fourth-largest Token supply platform with a market share of 1.5% in independent life..

On June 17, Mi officially released yesterday the official version of the cloud light quantification product Xiaomi Mimo Claw, pre-positioning the MiMo-V2.5-Pro flagship model for depths appropriate to OpenClaw framework, the Native Matching MCP tool call protocol, equipped with a million-scale super-long context, and the primary low-threshold experience of Mimo model and AI smart body capabilities. Officially, based on the MTP 3-tier decoded structure, the full-link job insulation efficiency increased by about 3 in OpenClaw standard intelligent workflow testing..

On June 16, Microsoft CEO Satya Nadra released a long-form post yesterday proposing the concept of “tokeen capital”, arguing that the competitiveness of AI-era firms will depend on the ability to build a learning gap between human capital and token capital. Each enterprise must build two types of capital at once: human capital and token capital. Human capital is the knowledge, judgement, interpersonal relationships, creativity and regular insight that employees possess; token capital is the AI capacity to build and manage an enterprise. He stressed, token..

On June 15th, Microsoft CEO Satya Nadra appeared at the New York Times technology podcast, Hard Fork, on the concept of the current fire. The 58-year-old Microsoft Master said, "You can't deny that people have a bad look at AI." He acknowledged that AI development could indeed lead to some of the manual jobs being replaced by machines, but that the enormous capabilities that AI has in place could provide many opportunities for human society. Nadra thinks everyone should be an AI stakeholder. AI can't just use one or two front-line models or a few..

On June 15, according to The Information, Meta issued an internal memorandum this Tuesday to approximately 60000 employees announcing that restrictions on the use of AI by employees would be imposed to curb the continuing rise in internal AI costs. The memorandum showed that Meta employees had accumulated 60.2 trillion tokens over the past 30 days, and that since then usage had climbed further to 73.7 trillion tokens, after which internal rankings had been dropped. The company expects that this year's internal AI usage expenditure will..

In recent days, the Director of the AI "API Transit Station" station in Shanghai has been on the Internet and has been detained for 37 days in criminal custody on suspicion of selling offshore AI interface resources, which has since been changed to bail and is still under investigation. As soon as news emerged, a large number of similar platforms were shut down and moved, and the whole industry was booming. At present, there are rumours that the case may be prosecuted for illegal dealing, so what is the space for defence? What's the business at the stop? A brief description of the operational modalities of such platforms. Because the main models of OpenAI, Claude and others have made land for mainland China..

On May 18, Chinese telecommunications yesterday launched a series of trial commercial Token packages for developers and small and medium-sized micro-enterprises, individuals and home-based users, providing "Token+connect + Secure" integration services, all of which offer three sets of food packages, which are priced at 9.9 million yuan/month for individual and household users; and 39.9 million yuan/month for developers and small and medium-sized micro-enterprises. Both types of users can choose to purchase broadband up speeds 50 Mbps, 100 yuan/month and smart body safety housekeepers..

According to news from 9 May, as reported yesterday, the data show that, by March this year, our daily average Token (words) called over $14 trillion, a surge of over 100000% and a thousand times the increase over two years. Since this year, it has been reported that a number of large domestic and foreign models have been connected to new and Agent's rapid spread, leading to continuous warming in the calculator market. According to experts, the current supply and demand situation is still expected to grow over time. The Deputy Director of the Advanced Computing Research Unit at Sadie Institute said that the current computing and leasing industry..

On 30 April, according to Stratechery, OpenAI CEO Sam Altman, in an interview with a tech commentator Ben Thompson, stated that the AI pricing model based on token would not be sustainable in the long term and that the industry would eventually shift to a pricing system based on “mission accomplished”. Altman exemplifies this judgement by using the latest GPT-5.5 model: the single token price of GPT-5.5 is higher than the previous generation GPT-5.4, but completes the same..

On April 30th, yesterday, AI researcher Aran Komatsuzaki published the results of a cross-evaluation of the main mainstream paraphrasing tool (tokenizer), which reveals "linguistic discrimination" in Tokenizer: When using the same model, non-English speakers actually consume far more tokens than English users, amounting to a quiet “non-English tax”. He translated Rich Sutton's famous thesis The Bitter Lesson into nine languages and divided..

On April 27th, OpenClaw was officially updated with 2026.4.24 versions, accessing the DeepSeek V4 series model, and DeepSeek V4-Flash, a default preferred model for new users. At the same time, DeepSeek officially updated the API file, announcing a downgrade to the price of DeepSeek's entire series of API services, entering a one-tenth drop in the Cache Price to the original price, Pro Model, May 5th, 2026..

In the morning of April 24th, OpenAI officially released GPT-5.5 and GPT-5 Pro, the most powerful flagship models under its flag. The core competencies are to accept a complex, multi-step, complex task and to do it autonomously - plan the path, call tools, validate the results, continue to advance without the need for gradual user intervention. Terminal-Bench 2.0 score 82.71 TP3T (GPT-5.4 for 75.11 TP3T), SWE-Bench Pro up to 5861 TP3T, internal ministerial mission assessment Expert-SWE..